EMDNA: Ensemble Meteorological Dataset for North America

Abstract:

Precipitation and temperature data are critical to hydrometeorological studies. However, quantitative estimates often have large uncertainties due to the inherent limitation of observing and modeling techniques including stations, radars, satellite sensors, and atmospheric models. To overcome this challenge, this study develops a probabilistic dataset, the Ensemble Meteorological Dataset for North America (EMDNA), which is at 0.1° and daily resolutions from 1979 to 2018 with 100 ensemble members. First, the Serially Complete Dataset for North America (SCDNA) is developed based on raw station observations from multiple sources. SCDNA includes daily precipitation, minimum temperature, and maximum temperature data for 27280 stations. Then, SCDNA and three reanalysis products (ERA5, JRA-55, and MERRA-2) are merged to provide requisite parameters for probabilistic estimation. Three major steps are implemented: (1) reanalysis data are re-gridded to 0.1°, corrected using linear scaling, and merged using Bayesian Model Averaging (BMA); (2) station regression estimates are obtained using locally weighted linear/logistic regression; and (3) reanalysis and regression estimates are merged using optimal interpolation (OI). Results show that OI-based merging achieves the most significant improvement in scarcely gauged regions such as northern Canada. Finally, the ensemble estimation is achieved based on the probability distributions defined by OI outputs and the spatiotemporally correlated random fields. EMNDA provides data for precipitation, mean temperature, and daily temperature range. It realizes uncertainty estimation of precipitation and temperature data and has a better representation of extreme events than deterministic datasets. Validation based on independent station observations shows that EMDNA achieves good performance in North America. In summary, EMDNA will be useful for a variety of research and applications such as ensemble hydrological modelling in North America.