The station data set from which this gridded data set has been constructed is an extension of the original CRU/US DOE data described in Eischeid et al. (1991). Substantial additional work in extending these station time series and increasing the network has been undertaken by the Climatic Research Unit in recent years. A total of over 11,880 station time series now exist. For access to these station data one should approach Russ Vose working on the Global Historical Climatology Network (GHCN Version 2) at Arizona State University, Arizona, US. (Email: rvose@smtpl.asu.edu)

Thiessen polygon weights were used to average gauge data within each gridbox. Where a monthly station value was missing an estimate was obtained by calculating the mean anomaly for that location derived from surrounding stations. This anomaly interpolation method required the station values to be converted into percentage anomalies from some reference period. These standard anomalies were then interpolated onto the missing station location using an inverse distance (with spherical adjustment), angular weighted method similar to that described in Shepherd (1984) and Legates and Willmott (1990). For this interpolation, a maximum percent anomaly value of 500 per cent was imposed. The interpolated percent anomaly was then converted back into a station mm estimate using that station's mean monthly precipitation total for the reference period. This mean anomaly interpolation was only performed for a missing station value where two stations within a 600km radius possessed valid data (for 1997 and 1998 this search radius was reduced to 400km to minimise instabilities in the resulting gridbox estimates). Otherwise the station value remained missing and hence the gridbox average could not be calculated for that month. A maximum of the 50 nearest stations could contribute to the interpolation.

(For 2.5 x 3.75 deg. data only)
Two gridded datasets were initially calculated based on two different reference periods: 1931-70 and 1951-90. Stations could only contribute to these climatologies if they possessed 75% or more valid monthly measurements in the reference period. For 1931-70, 5986 stations were used resulting in historical gridded time series for 1277 2.5° by 3.75° gridboxes. The maximum number of stations per gridbox was 51. For 1951-90, 6655 stations were included generating time series for 1418 2.5° by 3.75° gridboxes. The maximum number of stations per gridbox was 45. These two datasets were then combined using the common 1951-70 period to blend the data on the basis of mean monthly values and their variance. When merged, a total of 1520 gridboxes possessed time series of which 726 had complete data between 1900 and 1998. The period of most complete coverage was from 1952-75; the year 1959 possessed no missing data in any of the 1520 gridboxes.

(For 5 x 5 deg. data only)
Two gridded datasets were initially calculated based on two different reference periods: 1931-70 and 1951-90. Stations could only contribute to these climatologies if they possessed 75% or more valid monthly measurements in the reference period. For 1931-70, 5986 stations were used resulting in historical gridded time series for 699 5deg gridboxes. The maximum number of stations per gridbox was 91. For 1951-90, 6655 stations were included generating time series for 768 5deg gridboxes. The maximum number of stations per gridbox was 92. These two datasets were then combined using the common 1951-70 period to blend the data on the basis of mean monthly values and their variance. When merged, a total of 813 gridboxes possessed time series, of which 248 had complete data between 1900 and 1998. The period of most complete coverage was from 1952-67; the year 1959 possessed no missing data in any of the 813 gridboxes.

All station data have been screened for gross outliers and typographical errors using a number of semi-automated techniques. Owing to the large spatial variability of precipitation these methods, however, are not foolproof. The GHCN has considered and implemented further improvements to these screening methods (Easterling and Peterson, 1995; Easterling et al., 1996). No corrections for gauge undercatch have been applied to the station data (cf. Sevruk, 1982; Legates and Willmott, 1990). A spatially varying, but temporally constant, correction could be applied to the estimates derived from Legates and Willmott (1990), although this would not alter the trends in the data. Applying time-dependent corrections to gauge time series on a global scale is a gigantic undertaking which may well not be either feasible or justifiable.

A number of Northern Hemisphere high latitude time series contain inhomogeneities due to varying sensitivities to snowcatch of different gauge designs and mountings. These have been well documented for certain countries (e.g. Russia; Groisman et al., 1991; Scandinavian countries - the North Atlantic Climatological Dataset) and work is underway to "clean" other country datasets (e.g. Canada). Groisman's 'adjusted' data have now been added to the master station dataset used here, as has the NACD archive, but further improvements in the reliability of this gridded dataset over high latitudes will follow. For the present, the user should be cautious about the precise interpretation of high latitude precipitation trends outside Russia and Scandinavia, especially in winter.

No topographic weighting has been applied to the interpolation scheme. A number of different methods exist for incorporating the effects of topography on precipitation (e.g. the PRISM and AURELHY methods and the spline algorithms of Hutchinson, 1995). However, the dependence of precipitation anomalies on elevation is much smaller and more ambiguous. Since the method used here only interpolates anomalies, and not precipitation values themselves, excluding the effects of elevation is reasonable. There are, however, other problems associated with using precipitation anomalies in a gridding algorithm like this and these are discussed by Hulme and New (1997).

Detailed Summary taken from http://www.cru.uea.ac.uk/~mikeh/datasets/global/

Click on the links below to view data coverage maps for a particular time period. Percentage of non-missing data per time period is plotted.

(1901-1910, 1911-1920, 1921-1930)

(1931-1940, 1941-1950, 1951-1960)

(1961-1970, 1971-1980, 1981-1990)

(1991-1998)

Updated: 1/10/03
Maintained by asphilli@ucar.edu