The one of the objectives of the second phase of the Global Soil Wetness (GSWP-2) was targeted to produce the best estimate of global water cycles for years from 1986 through 1995. Offline simulation of the energy and water balance at land surface was calculated by more than 10 land surface models (LSMs), and sent to the Inter-comparison Center (ICC) of the GSWP2, founded at the Institute of Industrial Science, University of Tokyo. It is expected that simple ensemble mean or some statistical processing of the submitted outputs from the LSMs will provide the state-of-the-art information on the global energy and water balance.

Firstly, the water balance components of 10 year averaged annual mean value were compared with previous estimates. Global mean, continental mean, and zonal mean values of precipitation, evapotranspiration, and runoff were compared with historical publications, other datasets, and previous estimates. Global annual precipitation increased 17% compared to the forcing data used in the first phase of GSWP (GSWP-1). It is significant in Europe, where the increase is approximately 60%. Corresponding to this, annual runoff estimated by a simple Bucket model in Europe increased from 1,680 km3/y of GSWP-1 to 5,911 km3/y by GSWP-2. This value is extremely large relative to the estimates of 2,900 km3/y by WMO (1997) and 2,770 km3/y by the University of New Hampshire (1998). The forcing data of precipitation for the GSWP-2 was corrected considering the under catch by wind considering the rain gauge type in each country. It is suspected that this correction algorithm might have some deficiency since national borders can be seen in the precipitation data particularly in Northern Europe. It is possible that the original precipitation data in the corresponding European region had already corrected for wind under catch, and the GSWP-2 forcing data in the region was over-corrected. More examination and sensitivity study should be needed on this issue.

The similarity of the global water balance among the outputs by various LSMs are examined for annual mean field of runoff, annual amplitude of soil moisture in the root zone, the maximum soil depth, ratio of interception loss in the total evapotranspiration, and the surface runoff ratio. The misaligned outputs by a few LSMs were excluded from further data analysis, and weighted ensemble mean of outputs from remaining LSMs are estimated considering the similarity of the LSMs. Linear weighting confirms the conservation of mass in the water balance.

Finally, the obtained ensemble mean dataset is analyzed to present new insights in global water cycle. Global mean water balance explicitly including the interception-loss and transpiration, surface and sub-surface runoff, snow ratio in precipitation and snow accumulation, etc., is presented with their inter-annual variations in 10 years and uncertainties inferred from the variance among estimates of various LSMs. Budyko diagram, which adopts net-radiation over precipitation and evapotranspiration over precipitation in annual basis as for major variables to explain water and energy balance, is used in accordance with vegetation cover to classify the global lands into several climatic types. Representative water balance in these land use types is also presented.

Finally, the obtained natural water cycle is contrasted with socio economic information such as population and irrigated cropland area in each country, and analyzed in order to provide basic information for world water resources assessments.