Evapotranspiration and water budget (2003-2021) of an alpine shrubland on the northeastern Qinghai-Tibetan Plateau, China

Water provision ability is a pivotal ecological service of high-altitude alpine regions and is controlled by evapotranspiration (ET), precipitation, and soil water storage whereas the underlying ecohydrological processes remain highly unquantified. Since August 2002, ET has been continuously measured by an eddy covariance system, which comprised a 3-D ultrasonic anemometer (CAST3, Campbell Scientific, Logan, UT, USA) and an open-path infrared gas analyzer (Li7500, Li-Cor, Lincoln, NE, USA), mounted at a height of 2.5 m aboveground. The raw data frequency of 3-D wind speed, virtual temperature, and the density of CO2 and water vapor was uniform 10 Hz. The micro-climate was simultaneously observed by an adjacent auto weather station. These variables were mainly composed of 1.5 m air temperature and water vapor pressure (HMP45C, Vaisala, Finland), 1.5 m wind speed and direction (034A and 014A, RM-Young, USA), 1.5 m four-component radiation (CNR-4, Kipp&Zonen, Netherlands) and photosynthetically photon flux density (LI-190SB, Li-Cor, USA), 0.5 m precipitation (52202, RM-Young, USA), −0.05 m (a negative value means belowground) soil heat fluxes (HFT-3, Campbell Scientific, USA), −0.05 m soil temperature (105T, Campbell Scientific, USA), and −0.10 m and −0.20 m volumetric soil water content (CS616, Campbell Scientific, USA). The auto weather station was synchronized with the eddy covariance system and measured at 30-minute intervals. Because a tilting bucket precipitation gauge cannot accurately monitor snow and hail, monthly precipitation data were corrected by the corresponding observations of Haibei Station if necessary, which were measured manually two times (8:00 and 20:00, Beijing Standard Time) each day according to the conventional meteorological observation specifications.Satellite-derived normalized difference vegetation index (NDVI) was used to capture the temporal variations of plant growth and then to quantify the importance of vegetation to water balance. The 16-day NDVI was centered on the flux tower at a spatial resolution of 250 m and was generated by the Oak Ridge National Laboratory Distributed Active Archive Center (ORNL DAAC, http://daac.ornl.gov/MODIS/modis.html). Only the growing season NDVI (> 0.15) was used for further analyses, since alpine plants are dormant during the non-growing season. The abnormal values in NDVI serials probably caused by high cloud coverage were discarded and temporally interpolated. The 30-minute ET was recomputed from raw frequency data by EddyPro 7.0.4 (Li-Cor, USA). The main procedures included spike removal, 2-D coordinate rotation, time-lag compensation, Webb-Pearman-Leuning air density correction, low- and high-frequency spectral correction, and quality check. We discarded the poorest quality observations flagged “2” and kept the others (Foken et al., 2005). The latent heat flux dataset was filtered to remove outliers (<-100 W m-2 or >500 W m-2) and night-time friction velocity (> 0.15 m s-1) because of sensor malfunction or low turbulence. ET data gaps mainly occurred during the non-growing season and the nighttime and were gap-filled by thEvapotranspiration; soil water storage change; water budget; alpine shrublands; eddy covariance techniques; Qinghai-Tibetan Plateaue boosted regression trees model (BRT). BRTs of the growing season and non-growing season were separately trained by the valid data of ET and key environmental controls, including Ta, water vapor pressure, solar radiation, wind speed, soil temperature, and soil water content. The fitted BRT, combined with environmental controls corresponding to ET gaps, was used to fill those gaps.