Topographic Bias in Mesoscale Precipitation Networks

Abstract

In most weather station networks there are proportionately too few stations at high elevation, and areal estimates of climatological quantities that vary with elevation are biased. Two-dimensional interpolation between stations cannot remove this bias. The topographic bias can be understood, independently of any knowledge of quantities measured at the weather stations, if information is available on station elevations and on the ?real? topography. In particular, the bias can be broken down into two components: resolved bias, which is removable by generating an interpolated ?fictive topography? from the station topography, and unresolved bias, which remains even after interpolation and is measured by the difference between fictive and real topography. If resolved and unresolved biases differ in sign, the fictive topography is worse than the station topography as an estimate of the real topography. Biases were evaluated for 18 5° ? 5° blocks in the contiguous United States, using station elevations from a standard dataset and real elevations from a digital elevation model with ?8-km spatial resolution. The resolved biases in estimates of average elevation (station minus fictive) range from ?197 to +166 m. The unresolved biases (fictive minus real) range from ?398 to +10 m. The net biases (station minus real) range from ?389 to +18 m. Biases in elevation for blocks with higher relief are substantial. and several have magnitudes exceeding 10% of the magnitude of real elevation. Topographic bias in area] estimates of annual average precipitation was evaluated by fitting linear functions of location and elevation to precipitation data from weather stations. For 15 of the 18 blocks station precipitation was found (with 95% confidence) to increase linearly with station elevation. Fictive and real precipitation were calculated by substituting fictive and real elevation as arguments in these location-elevation functions. Resolved biases in precipitation are small and of variable sign. Unresolved biases in precipitation are also small in low-relief blocks with reasonably representative station topography, but all high-relief blocks have topographically biased networks; the unresolved bias averages ?13%, and the net bias ?11%, of real precipitation. These biases are comparable with other, better-understood biases in precipitation due to such causes as gauge undercatch, but the methods described here are capable both of identifying and, in principle, of correcting them. Estimates of other important variables, notably temperature, are also likely to he topographically biased in mountainous regions.