DEPENDENCE OF LARGE-SCALE PRECIPITATION CLIMATOLOGIES ON TEMPORAL ANDSPATIAL SAMPLING

Large-scale observed precipitation climatologies are needed for a vari ety of purposes in the fields of climate and environmental modeling. A lthough new satellite-derived precipitation estimates offer the prospe ct of near-global climatologies covering the last one or two decades, historical assessments of precipitation and its variability in time re main dependent on conventional gauge observations. A number of questio ns need to be asked of the existing precipitation climatologies that u se such gauge observations. What time period do they sample? What is t he spatial density of gauge coverage? What adjustments are made for me asurement bias? And what interpolation method is used to convert them to regular grids? Different precipitation climatologies, nominally des cribing the same variable, can yield very different answers when used as inputs in either the fields of climate model validation or environm ental modeling. This paper explores some of the reasons for these diff erences by examining the importance of the first two questions listed above-the temporal and spatial sampling of the precipitation normals t hat form the basis of these types of climatologies. The authors draw u pon subcontinental examples from tropical North Africa and Europe and show that, in the presence of significant decadal-scale precipitation variability, climatologies constructed from the same station network, but sampling different 30-yr time periods (i.e., 1931-60 and 1961-90), can vary by 25% or more. Using the same two regions, the authors also examine the influence of different spatiotemporal gauge sampling stra tegies on the construction of a long-term, ''twentieth-century,'' prec ipitation climatology. They show that, in the presence of multidecadal variability in precipitation, a strategy that favors more complete sp atial coverage at the expense of temporal fidelity can induce biases o f 5%-10% in the resulting climatology. They compare their 30-yr precip itation climatologies with those of Legates and Willmott and Leemans a nd Cramer. In Europe, where twentieth-century pracipitation exhibits l ittle interdecadal variability at the regional scale, different interp olation methods and station networks are the major cause of variations between these climatologies. Conversely, in tropical North Africa, wh ere historical precipitation shows decadal-scale departures from the l ong-term mean, differences between climatologies due to temporal sampl ing strategies are at least as great as those arising from alternative interpolation techniques and station distributions. The authors argue for careful consideration of the appropriateness of a given climatolo gy for any application, in particular the time period it represents, o r at least an awareness of potential pitfalls in its use.