An evaluation of gridded weather data sets for the purpose of estimating reference evapotranspiration in the United States

This study assessed the quality of gridded weather data for calculating reference evapotranspiration (ETref), which, by definition, represents a near maximum ET occurring in a well-watered agricultural environment. Six gridded weather data sets - GLDAS-1, NLDAS-2, the CFSv2 operational analysis, gridMET, RTMA, and NDFD were compared to weather data collected from 103 weather stations located in well-watered settings across the conterminous United States. ETref along with the weather variables used to compute it - near-surface air temperature, vapor pressure, wind speed, and shortwave solar radiation - were compared. The gridded weather data sets generally overestimated the standardized Penman-Monteith ETref produced from weather station data, with median biases ranging from 12 to 31 %. The overestimation was mainly due to chronic overstatement of air temperature, shortwave radiation, and wind speed and understatement of humidity. These results indicate that gridded data should be carefully evaluated before being substituted for agricultural weather station data. Bias correction procedures may make these gridded data more suitable for generating ETref. RTMA was generally the best performing gridded data set for all variables and NLDAS was the worst for all variables except vapor pressure. NDFD one-day forecasts outperformed most of the analysis products, likely due to its initialization with RTMA. gridMET temperature agreed relatively well with the station data due to its dependence on the PRISM station-interpolated data set. However, its performance was similar to its other parent product, NLDAS, for the remaining variables which reduced its ETref performance. The low-resolution products GLDAS and CFSv2 performed better than the finer resolution NLDAS product suggesting that spatial resolution is not a primary factor determining correspondence to station data. The excellent performance of gridMET temperature and RTMA indicates that the degree to which gridded data depend on station data is a primary factor determining correspondence.