Precipitation
This is a brief summary of methods used to assemble the GHCN-Monthly precipitation dataset. The methods were generally comparable to those used in developing the temperature dataset, so please consult the temperature methods section for additional information.
Duplicate Elimination
A precipitation time series for a given station can frequently be obtained from more than one source. For example, rainfall data for Beijing were available in three different source datasets. In brief, duplicate stations were identified by comparing each station with all other stations in all source datasets. Several statistics were used to describe the similarity between stations, including the number of identical months of data, the length of the longest run of identical months, and the number of identical values that were zero. These diagnostic statistics were used in conjunction with station metadata to subjectively determine whether stations were duplicates. In most cases, the decision was relatively straightforward, although a few degenerate time series posed proved more challenging.
Quality Control
A variety of tests were employed to assess data quality. The first step involved comparing stations with a gridded climatology and plotting the stations for visual inspection. Both of these processes uncovered mislocated stations and the former uncovered stations that were digitized 6 months out of phase. Additionally, each time series was tested for significant discontinuities using the Cumulative Sum test (which looks for changes in the mean) and an analogous test that looks for changes in the variance or scale. Each time series was also evaluated for runs of three or more months of the same nonzero value. Finally, each individual precipitation total was evaluated to determine if it was an outlier in space and/or time using a variety of nonparametric statistics.
