This is the second part of a series on benchmarking raw 1-h high-resolution numerical weather prediction surface-temperature forecasts from NOAA’s High-Resolution Rapid Refresh (HRRR) system. Such 1-h forecasts are commonly used to underpin the background for an hourly updated surface temperature analysis. The benchmark in this article was produced through a gridded statistical interpolation procedure using only surface observations and a diurnally, seasonally dependent gridded surface temperature climatology. The temporally varying climatologies were produced by synthesizing high-resolution monthly gridded climatologies of daily maximum and minimum temperatures over the contiguous United States with yearly and diurnally dependent estimates of the station-based climatologies of surface temperature. To produce a 1-h benchmark forecast, for a given hour of the day, say 0000 UTC, the gridded climatology was interpolated to station locations and then subtracted from the observations. These station anomalies were statistically interpolated to produce the 0000 UTC gridded anomaly. This anomaly pattern was continued for 1 h and added to the 0100 UTC gridded climatology to generate the 0100 UTC gridded benchmark forecast. The benchmark is thus a simple 1-h persistence of the analyzed deviations from the diurnally dependent climatology. Using a cross-validation procedure with July 2015 and August 2018 data, the gridded benchmark provided competitive, relatively unbiased 1-h surface temperature forecasts relative to the HRRR. Benchmark forecasts were lower in error and bias in 2015, but the HRRR system was highly competitive or better than the gridded benchmark in 2018. Implications of the benchmarking results are discussed, as well as potential applications of the simple benchmarking procedure to data assimilation.