Assessing the spatial representativeness of air temperature records

Monitoring long-term climatic change relies heavily on observational station records of near-surface air temperature. In the United States, these records are primarily drawn from the NOAA cooperative observer network. Stations included in climate analyses generally have long records with relatively few station moves, instrument changes, or other operational changes that impart discontinuities or degrade the observational record. With few exceptions, local environmental characteristics around weather stations have not been systematically analyzed or documented for their influence on spatial representativeness and have not been properly prioritized in station selection criteria.

One of the primary concerns of this paper was to determine if local environmental characteristics influenced historical station records to the extent that spatial representativeness was limited and that network homogeneity was compromised. Computationally intensive resampling procedures and geostatistical semivariance functions were used to iteratively test spatial representativeness of long-term weather stations. In addition to local environmental characteristics, this work also identified operational inhomogeneities that degraded spatial representativeness. Thirty-three stations (of 237) in the central United States subset of the Daily Historical Climatology Network were found to be spatially unrepresentative of regional-scale climatic variability. These stations deleteriously influenced the spatial homogeneity (patterns and averages) of long-term air temperature change. Since anomalous stations have even greater consequence in regional and local change assessments, they should be culled from or de-emphasized in historical climatology networks.