Soil moisture plays an important role in the water cycle within the surface ecosystem and it is the basic condition for the growth and development of plants. Currently, the spatial resolution of most soil moisture data from remote sensing ranges from ten to several tens of kilometres whilst those observed in situ and simulated for watershed hydrology, ecology, agriculture, weather and drought research are generally less than 1 kilometre. Therefore, the existing coarse resolution remotely sensed soil moisture data needs to be down-scaled. In this paper, a universal soil moisture downscaling model through stepwise regression with moving window suitable for large areas and multi temporal has been established. Datasets comprise land surface, brightness temperature, precipitation, soil and topographic parameters from high resolution data, and active/passive microwave remotely sensed soil moisture data from Essential Climate Variables (ECV_SM) with 25 km spatial resolution were used. With this model, a total of 288 soil moisture maps of 1 km resolution from the first ten-day of January 2003 to the last tenth-day of December 2010 were derived. The in situ observations were used to validate the down-scaled ECV_SM for different land cover and land use types and seasons. In addition, various errors comparative analysis was also carried out for the down-scaled ECV_SM and original one. In general, the down-scaled soil moisture for different land cover and land use types is consistent with the in situ observations. The accuracy is relatively high in autumn and winter. The validation results show that downscaled soil moisture can be improved not only on spatial resolution, but also on estimation accuracy.