Analysis of measured evapotranspiration shows that subsurface plant‐accessible water storage (PAWS) can sustain evapotranspiration through multiyear dry periods. Measurements at 25 flux tower sites in the semiarid western United States, distributed across five land cover types, show both resistance and vulnerability to multiyear dry periods. Average (±standard deviation) evapotranspiration ranged from 660 ± 230 mm yr⁻¹ (October–September) in evergreen needleleaf forests to 310 ± 200 mm yr⁻¹ in grasslands and shrublands. More than 52% of the annual evapotranspiration in Mediterranean climates is supported on average by seasonal drawdown of subsurface PAWS, versus 29% in monsoon‐influenced climates. Snowmelt replenishes dry‐season PAWS by as much as 20% at sites with significant seasonal snow accumulation but was insignificant at most sites. Evapotranspiration exceeded precipitation in more than half of the observation years at sites below 35°N. Annual evapotranspiration at non‐energy‐limited sites increased with precipitation, reaching a mean wet‐year evapotranspiration of 833 mm for evergreen needleleaf forests, 861 mm for mixed forests, 558 mm for woody savannas, 367 mm for grasslands, and 254 mm for shrublands. Thirteen sites experienced at least one multiyear dry period, when mean precipitation was more than one standard deviation below the historical mean. All vegetation types except evergreen needleleaf forests responded to multiyear dry periods by lowering evapotranspiration and/or significant year‐over‐year depletion of subsurface PAWS. Sites maintained wet‐year evapotranspiration rates for 8–33 months before attenuation, with a corresponding net PAWS drawdown of as much as 334 mm. Net drawdown at many sites continued until the dry period ended, resulting in an overall cumulative withdrawal of as much as 558 mm. Evergreen needleleaf forests maintained high evapotranspiration during multiyear dry periods with no apparent PAWS drawdown; these forests currently avoid drought but may prove vulnerable to longer and warmer dry periods that reduce snowpack storage and accelerate evapotranspiration.

Gap-filled evapotranspiration from eddy-covariance flux towers with co-located meteorological-station sourced precipitation (rain, snow and aggregated rain and snow) and temperature data. The study consisted of 25 flux-towers and 25 co-located or most-representative meteorological stations distributed across 6 states (California, Idaho, Colorado, Arizona, and New Mexico) in the semiarid western United States. Snow depth data were taken from either on-site snow-depth measurements or snow-pillow data, representative SNOTEL data or from the Coorperative Observer Network. Raw data (level 0, defined here as the non-gap-filled data obtained from Ameriflux)

Time period: water year 1998 through water year 2015.

Flux-tower provide continuous observations of energy, carbon and water fluxes at 15-min time intervals. Data were obtained through Ameriflux (https://ameriflux.lbl.gov/) and through direct correspondence with Flux-tower principal investigators. Co-located/representative meteorological stations were selected based on literature review and direct correspondence with flux-tower principal investigators. Sources of instrumentation description for each flux tower and data processing procedure is described in Rungee et al. (2018, https://doi.org/10.1002/hyp.13322).

Time zone: Pacific Standard Time (PST from 1/1/2006).

Level 0: 30-min Flux-tower data obtained from Ameriflux or flux-tower principle investigator, precipitation data from co-located or on-site meteorological station, and snow-water-equivalent or snow-depth data from co-located or onsite measurements. Detailed data sources are provided in Rungee et al. (2018, https://doi.org/10.1002/hyp.13322).

Level 1: Flux-tower data organized into .mat files (.csv files also) and ready for gapfilling. Daily PRISM precipitation and temperature data collected from the 800-m cell overlaying the flux-tower coordinates. Snow data organized processed to represent monthly melt. See Rungee et al. (2018, https://doi.org/10.1002/hyp.13322) for more detailed processing methods and the Level 1 README.txt

Level 2: model-ready, gap-filled daily evapotranspiration, precipitation (rain, snow and aggregated rain and snow), temperature, and potential evapotranspiraiton.

Spatial Information of flux-tower is placed in the director “GIS_data”.