Dependence of Large-Scale Precipitation Climatologies on Temporal and Spatial Sampling

Large-scale observed precipitation climatologies are needed for a variety of purposes in the fields of climate and environmental modeling. Although new satellite-derived precipitation estimates offer the prospect of near-global climatologies covering the last one or two decades, historical assessments of precipitation and its variability in time remain dependent on conventional gauge observations. A number of questions need to be asked of the existing precipitation climatologies that use such gauge observations. What time period do they sample? What is the spatial density of gauge coverage? What adjustments are made for measurement bias? And what interpolation method is used to convert them to regular grids? Different precipitation climatologies, nominally describing the same variable, can yield very different answers when used as inputs in either the fields of climate model validation or environmental modeling. This paper explores some of the reasons for these differences by examining the importance of the first two questions listed above—the temporal and spatial sampling of the precipitation normals that form the basis of these types of climatologies. The authors draw upon 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 strategies on the construction of a long-term, "twentieth-century," precipitation climatology. They show that, in the presence of multidecadal variability in precipitation, a strategy that favors more complete spatial coverage at the expense of temporal fidelity can induce biases of 5%–10% in the resulting climatology. They compare their 30-yr precipitation climatologies with those of and . In Europe, where twentieth-century precipitation exhibits little interdecadal variability at the regional scale, different interpolation methods and station networks are the major cause of variations between these climatologies. Conversely, in tropical North Africa, where historical precipitation shows decadal-scale departures from the long-term mean, differences between climatologies due to temporal sampling 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 climatology for any application, in particular the time period it represents, or at least an awareness of potential pitfalls in its use.