The GPCC
claims to use Shepards method in spherical coordinates (Spheremap) installed
at the GPCC by David Legates in order to interpolate station observations to
a regular 0.5 degree grid (e.g. Rudolf et al, 1992 or Schneider et al, 2010).
Let us assume that this is true. Shepards method uses a maximum of 10
neighbouring stations. The GPCC claims to use about 3,300 stations in Germany
where there are about 175 grid points and therefore ignores at least half of
the station e.g. in their long-term climatology (normals) and their full-data
product. Another feature of Shepards method is that it only uses stations
closer than 5km to a gridpoint if at least one station is that close. This is
in order to have a best representation of the climate at the gridpoint. The higher the available station density compared
to the number of gridpoints the more likely it is that stations get ignored
by this strategy just because others are closer to the gridpoints. In the
case of Germany the majority of gridpoints has stations closer than 5km to
them and therefore the majority of station observations gets ignored by
GPCC’s interpolation strategy. Fig. 1 shows that the interpolation strategy
of the GPCC ignores most of the stations in Germany. Though this effect is less pronounced in areas with lower station density, there is always the risk that a station is closer than 5km to a gridpoint and thus stations farther away get ignored. The GPCC has the largest collection of monthly precipitation observations. With an increasing number of stations in the GPCC data collection the effect described here becomes more and more important. In an ideal world with a very high station density only a few percent of the global area would not be ignored by the GPCC interpolation strategy. A more general discussion can be found here. |
Figure 1: Spatial distribution of precipitation stations in Germany (black dots, 3300 stations) and area around gridpoints used in case at least one station is closer than 5km (orange areas, 407 stations) and in case a maximum of 10 stations is used (480 stations). In Germany the majority of stations gets ignored by the misuse of Shepards Method. |
The GPCC is informed about the misuse of Shepards method for years now but prefers to ignore it.
Literature:
Rudolf B., H. Hauschild, M. Reiss, and U. Schneider, 1992: Die Berechnung der Gebietsniederschlaege im 2.5 Grad-Raster durch ein objektives Analyseverfahren. Meteorologische Zeitschrift, 1, 1, 32:50
Schneider U., A.
Becker, A. Meyer-Christoffer, M. Ziese, and B. Rudolf, 2010: Global
Precipitation Analysis Products of the GPCC. Published at http://gpcc.dwd.de
Shepard D., 1968: A
two-dimensional interpolation function for irregularly-spaced data. – Proc. 23rd
ACM Nat. Conf., Brandon/Systems Press, Princeton NJ, 517-524.