On the reliability of the U.S. surface temperature record