Sensitivity of secular trends in precipitation data to observational errors

There has been considerable interest in estimating secular trends in precipitation data in various regions of the world. It is therefore important to ascertain the manner in which errors of observation affect estimated trends. For this purpose we have compared trends at 1219 stations in the contiguous United States for two data sets: (a) original observations, also called raw observations, and (b) the observations, adjusted to compensate for suspected errors. The adjustments were made at the National Climate Data Center, Asheville (Quinlan et al., 1987;karl andWilliams, 1987), In order to focus on the effects of observational errors we attempted to avoid the effects of filling of missing data by limiting the analysis to the period 1940–1984 for which the number of missing values is much smaller than earlier periods. A least-square linear regression was performed on the raw and adjusted data for each station and the slopes of the fitted lines were compared. The comparison was made for monthly, seasonal and annual precipitation values.The results for annual precipitation showed that 23 percent of the stations have trends of opposite signs in the raw and adjusted data. The trends were identical in annual data at only 11 percent of the stations. When monthly data are combined to form seasonal and annual averages the magnitude of the difference between the slopes of the adjusted and the raw observations generally increases, indicating that the errors in the individual monthly observations are correlated. When the station data were averaged to obtain state-wide averages, the effects of the errors became less pronounced in most of the states. These results indicate that obtaining trends in precipitation from station data is a more difficult problem than has been realized.