Urban-Effect Correction to Improve Accuracy of Spatially Interpolated Temperature Estimates in Korea

Abstract

Gridded temperature data are frequently used to run ecological models at regional scales and are routinely generated by spatially interpolating point observations at synoptic weather stations. If synoptic stations are located in urbanized areas, observed temperature and the interpolated data could be contaminated by the urban heat island effect. Without an appropriate correction, temperature estimates over rural areas or forests might deviate significantly from the actual values. This study was conducted to remove the urban effects embedded in the interpolated surfaces of daily minimum temperature in South Korea, where most weather stations are located in urbanized or industrialized areas. To overcome the spatially discontinuous nature of the population statistics, urban land cover information at a 30 m ? 30 m resolution was used along with population data. A population density was calculated by dividing the population of a city by the number of urban pixels falling within the city boundary. Population-density values unique to each city were, in turn, assigned to all the urban pixels. Blocks of 3 ? 3 pixels were aggregated to form a ?digital population model? (DPM) on a 90 m ? 90 m grid spacing. Temperature estimation error from the existing interpolation scheme, which considers both distance and elevation effects, was obtained at 31 synoptic station locations in Korea each month. They were regressed on the population information at the same locations, expressed in DPMs smoothed at the radial extent of 0.5, 1.5, 2.5, 3.5, and 5.0 km. Selected regression equations were added to the widely used distance?altitude interpolation scheme. This new method was used to interpolate monthly normals of daily minimum temperature in South Korea for the 1971?2000 period. Cross validation showed approximately a 30% reduction in the estimation error over all months when compared with those by the best existing method.