A Generalized Approach of Three dimensional Variational Data Assimilation for Land Surface Air Temperature and Its Tests under Simplification Cases

A generalized approach of threedimensional variational data assimilation (G-3DVar) with 1month assimilation window for land surface air temperature data was developed. As preliminary tests on this approach, two simplified forms of background error covariance matrix (B matrix for short) were applied to generate a high spacetime resolution grid analysis field of land surface air temperature in January 2000 in China through incorporating the CRU monthly mean data into the NCEP daily reanalysis data based on an observation operator associating the daily data and monthly mean data. The CRU monthly mean land surface air temperature data have low temporal resolution, but higher spatial resolution and better approximation to the station observations on the Tibetan Plateau, while the NCEP daily reanalysis data have high temporal resolution, but lower spatial resolution and worse approximation to station observations in these areas. A comparison shows that the qualities of both the daily and monthly mean data of the generated grid analysis fields from both simplified forms are all improved, which are basically consistent with the station observations in the south eastern and central areas of China and well match the topography in the Tibetan Plateau，Xinjiang and so on, where the observation stations are sparse. It indicates that the G3DVar is a new efficient approach for preparing meteorological driving field for land ecological model or land surface model. Further comparison also shows that the analysis field from the simplified form with daily variance in the B matrix is better than that from the simplified without considering daily variance in the B matrix. It is an important basis for further improvement of assimilation performance of G-3DVar without any simplification, ie. covariance is considered in the B matrix.