Uncertainties in water availability estimates derived from multiple precipitation and evapotranspiration datasets and fusion methods for semi-arid and arid regions

Study region: Mongolia, an arid and semi-arid region. Study focus: Water availability (WA), calculated as precipitation minus actual evapotranspiration (AET), is difficult to estimate in drylands because the small residual can be sensitive to differences between input products. We evaluated 37 WA estimates, comprising 35 combinations of seven precipitation and five AET products and two estimates based on arithmetic average (AA) and Bayesian three-cornered hat (BTCH) fusion. We compared their temporal and spatial characteristics and evaluated their performance with observed and gridded runoff. New hydrological insights: Across the 37 estimates, the mean long-term annual WA was 39.5 ± 63.5 mm yr⁻¹ (mean ± SD), and trend slopes averaged −0.5 ± 0.8 mm yr⁻². The CV of long-term means was 160.8%, and the relative slope spread was 158.7%. Most estimates showed higher WA in northern Mongolia, although some displayed the opposite spatial gradient. Runoff agreement was weaker in southern arid basins and varied among individual combinations. ERA5_ERA5 showed comparatively strong runoff agreement, while AA and BTCH exhibited relatively consistent, intermediate performance across basins and significant correlations with gridded runoff over relatively large areas of Mongolia. Precipitation-product selection accounted for 74.5% of the combined normalized main-effect variation associated with precipitation and AET selection among the 35 individual combinations. These findings support careful input selection and regional evaluation of fusion-based WA estimates.