Land surface variables, such as soil moisture, are among the most important components of memory for the climate system. A more accurate and long time series of land surface data is very important for understanding of land surface-atmosphere interaction and for improving our ability to predict weather and climate. Thus the main effort of this work is to rerun the NOAH land surface model retroactively as far as possible: 1948-present, which is essentially a 'Land Reanalysis'.

As the first step of the above effort, a 51 year (1948-1998) 1/8 degree grid and hourly retroactive LDAS forcing data set was generated, including air temperature, air humidity, surface pressure, wind speed (U,V), surface downward shortwave radiation, surface downward longwave radiation from global reanalysis and precipitation from CPC (Higgins etc). Some unique procedures were used to prepare this hourly retroactive forcing data in order to make a homogeneous forcing data set in the required spatial and temporal resolutions. The model parameters and fixed fields are derived from existing high resolution vegetation, soil coverage and orography.

The land dataset from NOAH land surface model is generated at 1/8 degree LDAS grid and it consists of 8 energy balance components, with output saved at 3-hourly time resolution, and 15 other variables (i.e. water balance components at 4 levels, surface state variables etc.), with output in daily resolution for the period January 1948 through December 1998.

The preliminary results show that the hourly forcing data set is reasonably good, compared with the observations. At the time of this writing, a few test runs have been done and the Retroactive LDAS Run has been run for 5 years, starting from the first day of 1948. The outputs provide an improved soil moisture and more associated land surface variable data set, such as snowpack, surface fluxes etc which we never had before. Many interesting data validation, data analysis and model comparisons are underway. The retroactive run will also provide superior model consistent initial conditions for numerical predictions. These studies are important for understanding land memory processes, evaluating and developing new land surface models, and eventually improving our understanding weather and climate change.