The Japan Meteorological Agency (JMA) operates a non-hydorostatic model (JMANHM) with horizontal resolution of 5km for severe weather prediction 8 times in a day.This research is carried out in order to sophisticate the land surface model(LSM) of JMANHM, as a part of "Innovative Program of Climate Change Projection for the 21st Century" supported by MEXT.

The LSM of JMANHM is based on SiB (Sellers et al, 1986) and improved the process of snow and soil, which has been also developed for GSM, the operational global model at JMA.

In short-term forecasts, it is important to prepare initial conditions of LSM adequately. However, the observational soil moisture and snow water equivalence(SWE) are insufficient in the forecast domain around Japan.Therefore, fine land surface states in the forecast domain are estimated by driving the off-line LSM for some past years with observed data as forcing such as high resolution (1km x 1km) precipitation analysis (Radar-Rain Gauge Analyzed Precipitation), surface temperature, surface wind and sunshine duration.

The land surface data which is made by above method is utlized as initial conditions for JMANHM, and influences to the forecast of precipitation and land surface elements are evaluated.

Finally, off-line LSM is executed with above obeserved data in real time in order to estimate initial land surface data for the operational JMANHM.

We will mainly present the evaluation of the land sureface data by driving off-line LSM for past some years.