Assessment of Global Precipitation

 

A Project of the Global Energy and Water Cycle Experiment (GEWEX) Radiation Panel

GEWEX, World Climate Research Program, WMO

 

Lead Authors:

 

Arnold Gruber

Cooperative Institute for Climate Studies, Earth System Science Interdisciplinary Center, University of Maryland, College Park, Maryland

 and

Vincenzo Levizzani

Institute of Atmospheric Sciences and Climate, Italian National Research Council, Bologna Italy
Preface

 

William Rossow, Chair, GEWEX Radiation Panel

 

The charge given by the GEWEX Radiation Panel (GRP) to the Precipitation Assessment Group was to evaluate the reliability of available, global, long-term precipitation data products in depicting the variations of precipitation at larger than weather scales with a special emphasis on the Global Precipitation Climatology Project (GPCP) product that is produced under the auspices of the GRP. The original goal of GPCP was to produce a new precipitation product employing satellite observations that was globally complete and could provide a quantitative description of regional precipitation variations on seasonal to interannual time scales. However, the continuation of GPCP has allowed for extension of the record to 25 years (now approaching 28 years), so that the question of longer term variations arises. Moreover, a stronger emphasis on precipitation processes at weather scales has begun with GPCP efforts to obtain useful precipitation measurements on diurnal to daily time scales. The GRP views this assessment as one step in progress towards more accurate measurements of precipitation and as part of preparations for a systematic improvement, revision and re-processing of the global precipitation products.

 

Executive Summary

 

This assessment was conducted by an international group of scientist who are experts in the measurement and analysis of precipitation using remote sensing techniques and in situ gauges.  Although focused on the data set produced by the Global Precipitation Climatology Project, the assessment also reviewed the current state of the art of satellite techniques for estimating precipitation as well as the variety of long term gauge data sets.  An interesting aspect is that the satellite techniques discussed include single sensor (e.g., infrared, microwave)  as well as multi spectral techniques, and time scales less than monthly and space scales finer than 2.5 × 2.5 degrees latitude/longitude.  Clearly these retrieval algorithms are continuously evolving and we need to emphasize that the GPCP, which for the most part utilizes single sensor techniques that are decades old, will at some point have to consider the impact of new retrieval algorithms as well as sensors ( e.g. Tropical Rainfall Measuring Mission, TRMM).  This was suggested in Chapter 4 which called for a re-analysis where new retrieval techniques and sensors would be evaluated for use in global precipitation estimates along with higher space and time resolution data.

 

Chapter 3 provides and excellent review of the global mean precipitation and its spatial and temporal distribution. The analysis is based on the 25 year period 1979-2004, which exhibited a global mean of 2.61 mm day^-1.  With regard to this value the authors of Chapter 3 provide an estimate of the uncertainty of ± .03 mm day^-1.  They point out that at this level of uncertainty there is no significant mean annual cycle in global precipitation.  This is consistent with global energy arguments that to a first approximation global precipitation should be more or less constant over the 25 year period.  The mean annual cycle over the oceans and land are examined separately with the land areas showing the largest annual variation.  Analysis of the spatial and temporal distribution of precipitation demonstrated that this data set is very capable in capturing the ENSO, the major inter-annual variation in precipitation, most evident in the tropics but also influencing mid-latitude regions.  However, no relationship was found between global precipitation anomalies and ENSO.

 

The situation is not as clear with regard to longer period variations, especially since as noted in Chapter 3 that this data set was not designed for trend analysis.  Also, as noted in Chapter 3 the analysis indicated that there was no discernable trend in global averaged precipitation.  However, this does not preclude the existence of regional trends.  Analyses were presented that indicate small areas of linear trend over land and the Indian and central to eastern Pacific Oceans. Note that, however, these data seem to suggest that the rainfall shifts between the 1982/83 and 1997/98 ENSO. A similar result was obtained using an EOF analysis which isolated the ENSO regime (modes 1 and 2) from the lower frequency variations (mode 3).  Also, a recent analysis suggested that there were positive trends in the frequency of upper and lower amounts  of precipitation but compensated by a negative trend in the frequency of intermediate amounts.  Nevertheless, these trend calculations are very sensitive to the length of record and it was felt that with an increase in the GPCP record length questions concerning longer period variability and trends can be answered with greater confidence.

 

Based on the analysis presented in Chapter 3 we feel that it is crucial to continue this data set.  It is clearly useful for studying inter-annual variability and increasing the length of record would help increase the reliability in the low frequency changes calculated on a regional scale.  This would meet the requirements for applications of the data set to global climate analysis..

Chapter 4 provides a brief glimpse into the future.  Given the increase of new satellite retrieval algorithms and other gauge data sets it seems reasonable to anticipate that a re-analysis of the GPCP would take place that would be able to demonstrate an improved accuracy of the global precipitation.  One area in particular would be to try and utilize the TRMM precipitation radar data to provide an oceanic reference for ocean precipitation in a similar way that gauges provide for the land areas. 

 

Also identified was the need to determine snow rate using remotely sensed data and accurate precipitation in complex terrain, the latter being a problem for remote sensing techniques and gauges.  Another possibility in the future is the application of data assimilation methods to observed and modeled precipitation in order to obtain a dynamically, physically and hydrologically consistent field of precipitation.  This would require a collaborative research effort among data producers and modelers.

 

This chapter also identifies the international effort to obtain higher spatial and temporal resolution precipitation data through the Program to Evaluate High Resolution Precipitation Products (PEHRPP, http://essic.umd.edu/~msapiano/PEHRPP/) Project.  The creation of datasets in this direction will significantly enhance the usefulness of precipitation data from satellite sensors for regional climate analysis, which is a rapidly growing research area.

 

Finally the most significant future for global precipitation is the Global Precipitation Mission (GPM). Briefly, this will be a satellite mission that will consist of a core satellite with an advanced dual-frequency precipitation radar and microwave instruments and a constellation of polar orbiting satellites whose precipitation estimates can be calibrated against those of the core satellite.  It will extend the TRMM mission by providing coverage at higher latitudes at 3 hour intervals over nearly the entire globe.  Clearly a challenge facing the global precipitation community is to develop methodologies for utilizing these new observations to improve and extend existing data sets such as GPCP thus providing the long time records for assessing climate change signals.

Chapter 1.                   Introduction

 

There are only a limited number of global precipitation data sets available for study of the global water cycle and its climatic variations as, for example, called for by the Integrated Global Observing Strategy Partnership Water Cycle Theme (2000).  A widely available set of global precipitation data is the one produced by GPCP (Huffman et al. 1997; Adler et al. 2003).  Although comparisons of this data set have been done with other global precipitation data (Gruber et al. 2000; Yin et al.  2004) it has not been independently and thoroughly assessed in terms of how reliable it is in representing temporal and spatial variations of precipitation for climate change and water cycle studies.  This is crucial since a variety of satellite estimates of precipitation are employed in this data set as well as new methodologies for merging the satellite and gauge data.

 

At a planning workshop held in August 2004 at the Cooperative Institute for Climate Studies, University of Maryland it was decided to focus such an assessment on GPCP monthly mean data set (Huffman et al. 1997; Adler et al. 2003) with inclusion of other data sets as necessary.  GPCP is an international effort initiated in 1986 as a project of the World Climate Research Program (WCRP 1986) and it enjoys broad community support, one of the reasons for selecting this data set for assessment.  Subsequently, the GPCP was incorporated into the Radiation Panel of the Global Energy and Water Cycle Experiment (GEWEX) of the WCRP. It was formed to improve understanding of seasonal to inter-annual and longer term variability of the global hydrological cycle, determine the atmospheric latent heating rates needed for weather and climate prediction models, and provide an observational data set for model validation and initialization and other hydrological applications.  Its initial goal was to produce a ten year climatology of monthly global precipitation on a 2.5 º × 2.5 º latitude/longitude grid.  In recognition of the vast areas of the globe that are not sampled by gauges it was clear that the project would rely heavily on satellite estimates of precipitation which would be merged with rain gauges where available. The early years of the project were spent in organizing the various components of the project ( Xie and Arkin 1994) and going through the process of evaluating and selecting algorithms for the retrieval of precipitation from geostationary and polar orbiting satellites using visible (VIS)-infrared (IR) and passive microwave (PMW) observations (Xie and Arkin 1994; Ebert et al. 1996).  The first version of the GPCP merged satellite and gauge data set was produced in 1997 (Huffman et al. 1997).  This version revealed a markedly different view of global precipitation, especially over the oceans, than previously depicted by other climatologies that did not have the benefit of satellite observations (e.g., Jäger 1976; Legates and Wilmott 1990).  The initial success of the project led to an extension of the precipitation data set back in time to 1979 (Adler et al. 2003) providing a record of global monthly precipitation of 27 years long and continuing.  Given the length of the GPCP climatology and its global coverage this data set is ideally suited for studying the global water cycle and has the potential for detecting a precipitation based global climate change signal.

 

This assessment reviews the procedures and input data used to produce the GPCP data set, its spatial and temporal variability, the future outlook for new and improved data sets, and recommendations about the quality and use of these data for studying the climate.  While the assessment will focus on the GPCP data set, other sources of global precipitation data will be included as needed to help support the analyses and conclusions of this assessment. 

Chapter 2.                   Global Precipitation Data Sets

 

2.1         Introduction

 

There is a pressing requirement for adequate observation and estimation of precipitation on a global scale stemming primarily from the paucity of such information over the vast majority of the Earth’s surface.  Conventional precipitation data sets, collected by gauges and, more recently, radar, suffer from spatial heterogeneity which given the temporal and spatial variability of precipitation leads to problems concerning the representativeness of the existing measurements.

 

Historically, precipitation has been measured in collection vessels such as the rain gauge (primarily for liquid precipitation) or snow gauges (for frozen precipitation).  Such gauges provide the basis of long-term precipitation data sets and are generally deemed to be representative of the precipitation at the point of measurement.  However, a number of factors affect the accuracy of such gauge measurements, such as gauge design, precipitation phase (liquid or frozen), wind effects, evaporation/condensation, etc.  Furthermore, gauges do not provide a reliable spatial measurement of precipitation.  The global distribution of gauges is quite variable ranging from relatively dense gauge networks in the more developed countries to sparsely distributed gauges in less developed regions.  Over the oceans gauges are essentially non-existent, with only a few gauges located on islands and atolls.  The representativeness of the gauges is therefore extremely important: a ‘good’ gauge density of 20 gauges per 1 × 1 degree latitude/longitude box implies that an area of 500 km² is represented by a sample typically collected from about 150 cm²: the vast majority of the globe has much poorer sampling.  Surface morphology (relief, vegetation, etc) over land and island locations over the ocean lead to significant spatial inhomogeneity in the distribution of precipitation.  Furthermore, heterogeneities arise from the characteristics of precipitation: convective precipitation tends to be localized and short duration, making its measurement more difficult, while stratiform precipitation is typically larger-scale and longer-term.  However, precipitation totals observed at neighboring stations usually have part of their variance in common depending upon season and region.  At monthly time-scales even stations which are separated by hundreds of kilometers have on average about 50% of their precipitation variability in common.

 

The development of radar systems to measure precipitation has addressed some of the short-comings of the gauge data sets.  First, radar is capable of providing a spatial measurement of precipitation (up to a certain distance from the radar location, typically about 100 km) and, second, it can provide frequent samples.  Radar does however have a number of disadvantages.  The conversion of the signal backscatter into rain-rates is not exact; surface effects and melting precipitation lead to anomalous signals, and low-level precipitation may be missed due to the upward-refraction of the radar beam through the atmosphere.  Other issues include attenuation, beam blockage, beam-filling, and beam overshoot (e.g., Sauvageot 1994). Radar networks can however be usefully employed with cross-calibration and calibration from gauge data, although in terms of global coverage, radars generally cover regions that already have adequate gauge networks.  The most useful application of radar in the generation of global precipitation datasets has probably been in the calibration and validation of satellite precipitation algorithms.

 

The estimation of precipitation on a global scale is therefore only viable through the utilization of Earth observation satellites.  The first meteorological satellite was launched in 1960 and since then a plethora of sensors have been developed and launched to observe the atmosphere.  These sensors fall into two main categories: VIS/IR sensors available from geostationary (GEO) and low-Earth orbiting (LEO) satellites and microwave sensors, currently only available from LEO satellites.  The suite of geostationary satellites is able to continuously monitor the Earth, providing data up to every 15 minutes in operational mode.  Meanwhile, the LEO satellites are capable of providing higher resolution data in the VIS and IR spectra, but only periodically when the satellites are passing overhead.  PMW data, collected from LEO, has much poorer spatial resolution than the VIS/IR measurements coupled with poorer temporal sampling associated with LEO observations.  Figure 2.1 and Table 2.1 provide an outline of the key satellites currently utilized for the retrieval of precipitation.

 

A range of algorithms and techniques has evolved to provide estimates of precipitation from the data collected by these sensors.  Estimates of precipitation derived from VIS/IR data sets rely upon the characteristics of the cloud tops: reflected VIS radiation can be used to infer the cloud thickness and height, while emitted thermal IR radiation is used to measure the temperature of the cloud tops.  Since all precipitation falls from clouds, the delineation of clouds themselves can provide a crude map of precipitation.  Algorithms such as the Geostationary Operational Environmental Satellite (GOES) Precipitation Index (GPI) described by Arkin and Meisner (1987) have shown that, despite their simplicity, over large space and time scales such techniques work reasonably well.  The availability of geostationary data at relatively high spatial and temporal scales permits the evolution of clouds systems to be studied and precipitation estimates to be generated.  Techniques such as the Griffith-Woodley technique (Griffith et al. 1978) and the convective-stratiform technique (CST, Adler and Negri 1987) exploit such data.  Operational techniques such as the Interactive Flash Flood Analyzer (IFFA, Scofield 1987) and subsequently the Autoestimator (Vicente et al. 1998, 2002) have been implemented to provide estimates of precipitation in real-time for a number of applications.  More recently, neural network techniques have been applied to these data sets like the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) (Sorooshian et al. 2000).

 

The primary drawback of the VIS/IR techniques is that the observations only relate to the characteristics of the cloud tops, rather than the precipitation reaching the surface. In the mid-1970s work on identifying precipitation from PMW observations showed much promise (e.g., Savage and Weinman 1975; Weinman and Guetter 1977).  Observations at microwave frequencies relate to the amount of water within the vertical column of the atmosphere being observed.  At frequencies below 40 GHz the precipitation signal is primarily due to the emission of radiation from precipitation-sized particles, adding to the upwelling radiation stream from the surface.  Above 40 GHz these particles start to scatter the upwelling surface radiation resulting in a reduction in the sensor-received radiation.  The former (emission) characteristics are best viewed over a radiometrically cold surface, such as over bodies of water, whilst the latter (scattering) are best seen over radiometrically warm surfaces, such as the land surfaces.

 

Many PMW techniques now exist for estimating rainfall, ranging from the relatively simple, empirically derived and calibrated techniques (e.g., Ferraro 1997), through to those that use complex atmospheric physics and radiative transfer equations to derive estimates of precipitation (Kummerow et al. 2001).  Comparisons between VIS/IR techniques and PMW techniques have shown that the PMW technique provides much better instantaneous estimates of precipitation (Ebert et al. 1996).  This is primarily due to the more direct nature of the observations. However, for longer-term estimates, the VIS/IR techniques based on geosynchronous data tend to perform better due to their better temporal sampling.

 

The combination of both the PMW observations and the VIS/IR observations has therefore been the subject of much work in recent years.  Adler et al. (1994) used PMW estimates to calibrate the IR precipitation estimates on large spatial and temporal scales.  More recently techniques to generate PMW calibrated estimates at high resolutions (on the order of 10 km / 30 minutes) have been devised (e.g., Turk et al. 2000; Kidd et al. 2003; Joyce et al. 2004; Huffman et al. 2006).  However, these techniques have yet to reach maturity and long time series of global precipitation estimates from these algorithms are not yet available.

 

 

2.2         GPCP monthly mean precipitation products

The GPCP is a mature global precipitation product that uses multiple sources of observations, including surface information. Huffman et al. (1995, 1997) describe the GPCP product generating estimates at the 2.5 × 2.5 degree monthly resolution, this resolution being later improved to 1 × 1 degree daily estimates (Huffman et al. 2001) and 2.5 × 2.5 degree pentad estimates (Xie et al. 2003). The current GPCP Version 2 Satellite-Gauge (SG) product is described here.

One of the major goals of the GPCP is to develop global precipitation analyses at monthly and finer time scales to permit a more complete understanding of the spatial and temporal patterns of global precipitation.  The merging of estimates from multiple sources takes advantage of the strengths offered by each type: local unbiased estimates where rain gauge data are available, physically-based PMW rain rates estimated from LEO satellites, and high temporal resolution indirect estimates from VIS/IR sensors on GEO satellites.  Data from over 6000 rain gauge stations together with satellite IR and PMW observations have been merged to estimate monthly rainfall on a 2.5 degree global grid from 1979 to the present.  The GPCP's Global Precipitation Climatology Centre (GPCC) maintains a collection of high quality rain gauge measurements that are used to prepare comprehensive land-based rainfall analyses.  The careful combination of satellite-based rainfall estimates provides the most complete analysis of rainfall available to date over the global oceans, and adds necessary spatial detail and bias reduction to the rainfall analyses over land. In addition to the combination of these data sets, careful examination of the uncertainties in the rainfall analysis is provided as part of the GPCP products.

2.2.1       Input data and characteristics

2.2.1.1            Gauges

 

For the period 1986 to the present the monthly gauge analyses are constructed by the GPCC operated by the German Weather Service.  The GPCC uses a variant of the spherical-coordinate adaptation of Shepard's method (Willmott et al. 1985) to interpolate the data observed at gauge stations to regular grid points at a resolution of 0.5 × 0.5 degrees.  These regular points are then averaged to provide monthly precipitation totals at the final 2.5 × 2.5 degree resolution.  This methodology helps counteract the