Systematic gage network reporting errors: Missing and Unreported Observations

In addition to true instrument-related errors, precipitation gage reporting networks are subject to systematic errors caused by inconsistencies in reporting. Examples include manual sites that report only on days with non-zero (or large non-zero) rainfall, automated sites with telemetering or data transmission problems resulting in regular periods of missing data, and sites with periods of incorrect constant rainfall. Although many of these problems do not result in clearly erroneous data, the use of affected data in gridded analyses, model verification studies, and other types of statistical sudies can lead to subtle but occasionally large inaccuracies.

We use hourly and daily reports from the realtime reporting gage network over the continental U.S. to demonstrate these effects. Where possible, the resulting impact is assessed by comparison with archived values from the high-quality retrospective Hourly Precipitation Dataset (HPD). We find that (1) many of the realtime stations have unrealistically large frequencies of daily precipitation; (2) area-averaged precipitation estimated from these stations during sporadic convective rain episodes are often 50% or more too large; and (3) automated data occasionally show nonphysical periodicities that appear to be caused by faulty telemetry. Methods to identify and correct for 'missing zero' stations are discussed. By examining distributions of daily rainfall rates observed at hourly stations, we set reasonable limits for hourly data dropout rates. Finally, to roughly estimate the effect of data problems such as these on model verification credibility, we present verification scores (including bias and equitable threat score) for Eta model runs computed with corrected and uncorrected gage reports.