LDAS Retrospective Analyses: A hydrological perspective

Accurate specification of soil moisture has important implications for numerical weather prediction. Particularly in summer, soil moisture exerts a strong control on evaporation, which in turn affects atmospheric heat and moisture budgets. A difficulty in the application of current-generation land surface schemes in numerical weather prediction models is specification of the soil moisture at the beginning of the forecast period. Soil moisture exhibits long memory relative to most atmospheric processes, so current approaches that use analysis fields to force the land surface scheme up to the forecast time are prone to accumulation of errors in the soil moisture. The Land Data Assimilation System (LDAS) avoids this problem by using observed surface forcings for a suitable period (typically months to years) prior to the forecast time to produce initial model soil moisture at the time of forecast. Several companion papers describe the LDAS project, and accomplishments to date with respect to forecast initialization.

Beyond the immediate implications for numerical weather forecasting, the LDAS project has spawned various ancillary analyses that are facilitated by LDAS data sets. Parallel to the real-time forecast initialization pathway in LDAS is a retrospective analysis pathway. This pathway involves long-term simulations using the LDAS land surface schemes, running over a period of multiple decades. The Variable Infiltration Capacity (VIC) model, in particular, has been run using LDAS surface forcings (gridded station data, including precipitation, temperature, humidity, surface pressure, and wind) for the period 1950-97, at the LDAS 1/8° resolution. These retrospective simulations serve two purposes. First, through use of observed streamflow data and implementation of parameter estimation methods, retrospective analyses allow identification of model parameters that cannot be determined directly from first principles or land cover data sets. Such parameters include, for instance, the VIC variable infiltration parameter b, which controls the partitioning of precipitation into infiltration and direct runoff. Second, retrospective analyses afford the opportunity to determine long-term excursions in land surface state variables, such as total subsurface moisture storage, as well as spatial patterns of surface fluxes (e.g., evapotranspiration). These quantities can, in some cases, be compared with values independently derived form atmospheric budgets. As such, they provide a check of diagnostic studies which previously have been based primarily on atmospheric analysis and reanalysis. In the VIC-LDAS retrospective analysis, surface forcing data (daily precipitation, maximum and minimum temperatures) are obtained from NCDC Summary of the Day data, wind from reanalysis, and solar and longwave radiation is derived from relationships with diurnal temperature range. Model soil and vegetation parameters and topographic information are obtained from the NRCS States Soil Geographic Data Base (STATSGO), the USGS EROS Data Center North America Land Cover Characteristics Data Base, and USGS digital topographic data, respectively. The model has been applied to the Arkansas-Red, Upper Mississippi, Missouri, and Ohio River basins, which constitute most of the Mississippi River basin. Model performance in terms of streamflow simulation is illustrated, as well as comparisons of soil moisture excursions over the period of analysis, and patterns of long-term surface heat and moisture fluxes, for the Mississippi River basin, as well as preliminary results for the continental U.S. based on extrapolation of VIC model parameters from the Mississippi.