The data available for estimating the global record of precipitation exist in a diverse and growing set of sensor-specific sources, each with particular periods of record, areal coverages, and algorithm strengths and weaknesses. Typically algorithm performance depends on the underlying surface characteristics and the weather regime that are associated with each observation. Consequently, a combination of the various data sources is needed to provide long-term global analyses of precipitation. Key issues in this work include accounting for asynoptic sampling, eliminating calibration differences among satellites, establishing absolute calibration, estimating errors, and choosing a period of record that maximizes data content and length of record. The Global Precipitation Climatology Project (GPCP) monthly precipitation product, initiated over a decade ago, is based on sequentially combining satellite estimates and using raingauge data where available. Subsequent versions of the GPCP monthly and daily products, and the Tropical Rainfall Measuring Mission (TRMM) three-hourly (originally daily) and monthly products continue this heritage, even while introducing new intercalibration schemes to support the use of additional data sources, such as the Advanced Microwave Scanning Radiometer for Earth Observation Satellite (AMSR-E) and the Advanced Microwave Sounding Unit B (AMSU-B). Particularly at smaller time and space scales, the intermittent, highly non-gaussian nature of precipitation works against easy application of optimum interpolation schemes that may be applied to continuous fields, such as sea surface temperature, and leads us to continue use of sequential calibration/combination approaches.

We have also found it important to pursue and encourage precipitation estimation from sensors not originally intended for this purpose. This work started with the incorporation of estimates inferred from Television Infrared Operational Satellite (TIROS) Operational Vertical Sounder (TOVS) data in Version 2 of the GPCP monthly product, as well as in the GPCP daily product. Subsequently the TOVS has been superseded by similarly computed estimates from Advanced Infrared Sounder (AIRS), which flies on the Aqua satellite. Such estimates are more approximate than those from precipitation-oriented microwave sensors, but provide critical information in places where current microwave sensors cannot provide estimates, such as regions of cold, frozen, and icy surfaces.

Continued introduction of new satellite sensors, research into combination approaches, and planning for the Global Precipitation Measurement (GPM) mission have accelerated the development of additional combination schemes. In the short term, we are seeking to push fine-scale estimates (i.e., three-hourly, 25-kilometer) from the current tropical/extratropical belt to the entire globe using precipitation estimates based on AIRS and other sounder data. Even after the introduction of high-frequency microwave sensors, we believe that the more approximate sounder-based estimates will continue to have a place in filling in holes in the higher-quality estimates comparable to the role of infrared-based estimates in the tropics today. Other possible sources of precipitation estimates include the high-frequency microwave sensors, as well as scatterometers. Another high priority in the near term is to develop defensible error estimates for the fine-scale estimates, including methods for estimating how the error changes with time/space averaging. We also plan to examine the extent to which our current paradigm of three-hourly snapshots needs to move to a shorter interval of, say, one hour. Such a change might require some form of interpolation to ensure an adequate supply of high-quality precipitation estimates.

One critical question for the climate community is the degree to which the community will push to have new-generation precipitation algorithms applied to archived data. For example, the GPCP is seeking to develop Version 3, and it is an open question how much effort should be expended on upgrading the input data and corresponding analysis algorithms to provide finer scale estimates than are currently available for archived data (2.5x2.5-degree monthly before 1997).