A principal goal of the GEWEX Continental-Scale International Project (GCIP) Program is to improve the analysis of precipitation over a range of space and time scales. Over the past several years the Climate Prediction Center (CPC) has developed a US Precipitation Quality Control (QC) System and Analysis that addresses two principal aspects of this goal: 1) Improved QC of raingauge data used in precipitation analyses for the United States and 2) Improved precipitation products and applications in support of climate monitoring, climate prediction, and applied research. Specific topics covered by this Atlas include:
1. Development of the U.S. precipitation QC system and analysis;
2. QC Initiatives involving radar and satellite data;
3. Daily, monthly and seasonal precipitation products / applications;
4. A Unified Raingauge Dataset (URD) for the U.S. (1948-present);
5. A daily precipitation reanalysis for the U.S. based on the URD.
The variability of U.S. precipitation at time scales ranging from daily to interannual is examined using the precipitation reanalysis. Comparisons are made to an existing climatology in use at CPC. An electronic version of this Atlas is found at the URL:
http://www.cpc.ncep.noaa.gov/products/outreach/atlas.html
Accurate and complete estimates of precipitation are critical to a wide variety of problems ranging from understanding the water budget to improved monitoring and prediction of climate. Most areas of the globe are not adequately sampled, either by in situ or remote sensing. The conterminous U.S. is covered by a relatively dense array of in situ (hourly and daily) raingauge data. Precipitation over the U.S. can also be estimated using satellite data and radar data that is archived at high temporal and spatial resolution. These resources allow us to focus on improving the quality of the analysis of precipitation in the U.S. over a range of space and time scales.
Improving the analysis of precipitation requires careful consideration of the quality of the input observations. In general the QC of gauge precipitation analyses has not been emphasized enough. The CPC routinely produces quality controlled gauge-only precipitation analyses for the U.S. as part of its effort to monitor current and past conditions and to provide improved climate forecasts for the U.S.. The major objectives of our activities have been to:
(1) Develop a near-real-time U.S. precipitation QC system and analysis;
(2) Improve the QC of gauge precipitation analyses using radar data and satellite data;
(3) Develop improved precipitation products and applications in support of climate monitoring, climate prediction and applied research;
(4) Compile a Unified Raingauge Dataset (URD) for the U.S. (1948-present);
(5) Complete a daily precipitation reanalysis for the U.S. based on the URD.
We note that the data sets in (4) and (5) are continuously updated as data becomes available. A flow chart that summarizes how U.S. raingauge data is processed at CPC is shown in Fig. 1. This Atlas is a summary of our accomplishments towards each of these objectives.
2.0 Near-Real-Time Daily Precipitation Analysis
The spatial coverage and accuracy of precipitation observations by "first order" stations in the U.S. is decreasing. Many in the climate community fear that the continued deterioration in surface observations will jeopardize our ability to perform real-time climate monitoring, forecasting and forecast verification. The problem is compounded by emerging requirements for daily (and even hourly) precipitation analyses.
There are several potential sources of precipitation data that could be used for precipitation analyses including 24-hr "first order" WMO GTS sites (near-real-time), 24-hr SHEF-encoded precipitation reports received via AFOS from the River Forecast Centers (near-real-time), hourly GOES/DCP and CADAS precipitation reports (near-real-time), hourly and 24-hr NCDC cooperative reports (non-real-time), and perhaps many other sources (e.g. SNOTEL data, HADS data). In December of 1996 the CPC organized a Precipitation Working Group to examine this problem in detail, in particular to inventory potential sources of suitable near-real-time precipitation data and to make appropriate intercomparisons to address issues of spatial coverage, reliability, and availability.
The group recommended the development of a near-real-time "U.S. Precipitation Quality Control (QC) System and Analysis" whose input was raingauge data. Such a system was built in early 1997 and has been undergoing continuous development and improvement since that time. A fully automated script was implemented to control data acquisition, run the QC (see section 3.0), prepare the analysis, archive the data and disseminate analysis products on the CPC Web Site (http://www.cpc.ncep.noaa.gov). Current products include daily accumulated precipitation, monthly and seasonal precipitation monitoring products, forecast verification products and drought / flood potential products (see section 4.1). The current suite of precipitation products for the U.S. also supports ongoing efforts in the CPC to deliver Climate Services, including the US Threats Assessment, a US Drought Forecast System, the Palmer Drought Index, and a Soil Moisture Forecasting System (see section 4.2). Our products are also used by many external research projects, including the Land Data Assimilation System (LDAS), and the NCEP Regional Reanalysis Project (see section 4.2).
2.2 Characteristics of the Daily Analysis
The daily analyses are gridded at a horizontal resolution of (lat,lon)=(0.25o x0.25o ) over the domain 140o W - 60o W, 20o N - 60o N using a Cressman (1959) scheme with modifications (Glahn et al. 1985; Charba et al. 1992). An intercomparison of precipitation analyses produced by Cressman (Cressmann 1959), Barnes (Barnes 1964), Shepard (Shepard 1968) and OI (Gandin 1963) schemes (not shown) revealed only minor differences in the analyses, presumably due to sufficient data density over the U.S. The input dataset for the near-real-time analysis is the CPC Cooperative dataset (24-hr "first order" WMO GTS sites and 24-hr SHEF-encoded precipitation reports received via AFOS from the River Forecast Centers). The analysis on Day 1 is valid for the 24-hour window from 1200Z on day 0 to 1200Z on day 1; a typical station distribution and daily precipitation analysis are shown in Figs. 2a and 2b respectively.
Several types of QC are currently applied to the gauge data: (1) A "duplicate station check" which eliminates duplicates and key punch errors from the raingauge reports; (2) A "buddy check" to eliminate extreme values; (3) a standard deviation check, which compares the daily raingauge data against a gridded daily climatology; and (4) NEXRAD radar QC of the gauge data to eliminate spurious zeroes; some details of the first 3 of these are discussed in section 3.1 while the fourth one is discussed in section 3.2. In addition, a fifth type of QC is included in the operational analysis involving satellite QC of the NEXRAD data (see section 3.3). All QC flags are inserted back into the gauge data archive for future reference. Station dictionaries are updated routinely to ensure proper elimination of duplicates as part of the QC procedure. It is anticipated that these QC initiatives will ultimately benefit radar-only and multi-sensor analyses.
Currently the daily analysis is available within ~16 hrs of real time.
3.0 Quality Control Initiatives
While the raw raingauge datasets are undergoing continuous development and improvement, there are nevertheless many problems with the resulting precipitation analysis despite the QC steps already in place; these problems are due to a combination of instrument error, bad raingauge reports that remain undetected and errors in the analysis scheme. High resolution radar and satellite-derived precipitation estimates offer potential for additional improvements to the QC of raingauge data.
There are three standard QC steps currently applied in our analysis system: (1) A "duplicate station check" which eliminates duplicates and key punch errors from the raingauge reports; (2) A "buddy check" to eliminate extreme values from the dataset and (3) a standard deviation check, which compares the daily raingauge data against a gridded daily climatology. The "buddy check" examines the absolute value of the difference between the current station and all stations within a one-degree grid box. If more than 50% exceed a specified threshold, then the current station is tossed. For the standard deviation check we currently use a daily climatology derived from the Unified Raingauge Dataset (see sections 5 and 6). The observations are compared to the nearest gridpoint value from the climatology. The current observation must be within 5 standard deviations (10 for hurricane events) of the daily climatology.
One serious problem in the CPC Cooperative Dataset is the number of incorrect reports of zero precipitation in the 24-hour SHEF-encoded "RFC" precipitation data (~6000-7000 reports daily). This problem is illustrated in Fig. 3, which shows the number of stations in the southeastern U.S. with no precipitation, less than 2 inches of precipitation, less than 4 inches of precipitation and greater than ten inches of precipitation for the period January-March 1998. The 1997/1998 El Niño event was characterized by heavy rain in the southeast US during this period, so it is clear that stations reporting no precipitation are in error.
While reports with erroneous large values are easy to detect and eliminate (i.e. via extreme value checks, buddy checks, etc), reports with erroneous zero (or small) values are hard to detect. A solution (currently implemented in our analysis system) is to eliminate spurious zeros from the raingauge data prior to analysis using hourly radar estimates of precipitation. One advantage of such a QC step is that it counters the tendency for underestimating the observed rainfall as is typically the case in gridded analyses.
The approach is as follows:
(1) Accumulate hourly radar precipitation estimates to 24-hr values (1200Z-1200Z).
(2) Compare all daily raingauge reports against the nearest gridpoint in the 24-hr radar estimate of precipitation (technically similar to the procedure used in our standard deviation check ). The high horizontal resolution of the radar data (4-km) works to our advantage since it ensures that the radar estimate and raingauge report are reasonably close to each other (i.e. within about 2-km).
(3) Eliminate raingauge reports below suitable thresholds in the radar estimates. A careful examination of the bias in the radar data (Fig. 4) suggested that a threshold of 2 mm day-1was suitable; this value is currently implemented in our analysis system.
(4) Insert the QC flags back into the CPC Cooperative Dataset. The QC information can be used to investigate the origin of these reports.
This QC step incorporates satellite based estimates of precipitation into the QC System. Radar estimates of precipitation are biased due to radar-radar calibration differences (when a single Z-R relationship is used), differences in precipitation rate between the radar scan level and the ground, and anomalous propagation of the beam (Fig. 4 shows two examples). In the past the QC of radar data has often been performed with information from other sensors (i.e. raingauge, satellite) and a number of investigators continue to examine this (Smith et al. 1997; Fulton et al. 1997; Seo et al. 1997; Ahnert et al. 1986; Hudlow et al. 1989; Office of Hydrology 1992). Recently, we developed an algorithm that uses satellite data to remove bias in radar estimates of precipitation before other QC steps are invoked. This algorithm was developed by porting software / experience from an earlier study by Joyce et al. (1998) into our QC system.
Basically, the algorithm uses high resolution GOES-IR data (currently 1/2 hour on a 1/2o x 1/2o latitude-longitude grid) to screen out heavy hourly radar precipitation estimates when collocated IR temperatures before, during and after the hour in question are warmer than a set threshold.
The closest 4 km GOES 8 or GOES 10 IR pixel is collocated with the midpoint of each 0.1o (lat,lon) hourly radar precipitation estimate using 30 minute IR images. The coldest pixel is determined from all three IR images for 3 spatial extensions: (1) the exact IR pixel collocated to the radar estimate; (2) all pixels within 25 km of the collocation; and (3) all pixels within 50 km of the collocation. The statistics are further separated by stratifying the collocations into categories of radar precipitation from 0 to > 25 mm hr-1 for classes every 5 mm hr-1 . The mean of the coldest IR pixel found is computed for all radar rainfall cases and the three spatial extensions. Standard deviation of the coldest IR pixel about the mean for each rainfall class is then computed, from which frequency maps of occurrences in classes of 0.5 sigma from the mean are computed. The mean coldest IR temperature (within 25 km of the radar estimate) for the class of no radar precipitation was the warmest at 262.9 K (Fig. 5a) with a standard deviation of 22.2 K (Fig. 5b). This quickly drops to 230.2 K with a standard deviation of 15.9 K for a radar rainfall of 0-5 mm hr-1. For cases of radar rainfall > 25 mm hr-1 the mean coldest IR temperature was 208.2 K with a standard deviation of 8.2 K.
The distribution of the cases of coldest IR pixel about the mean coldest IR pixel (Figs. 6a and 6b) reveals that for 80% of the no radar rainfall cases (May 1999) the coldest IR pixel is warmer than 240.0 K, or one standard deviation below the mean of 262.9 K. In almost 100% of the cases of radar rainfall greater than 25 mm hr-1, the coldest IR pixel is colder than 237.0 K, or 3.5 standard deviations above the mean (Figs. 6c and 6d), 240.0 K for rainfall greater than 20 (but less than 25) mm hr-1 . This gives considerable utility in eliminating incorrect radar estimates.
From the statistics previously described, an IR temperature threshold of 3.5 standard deviations warmer than the mean coldest IR pixel (within 25 km of the radar estimate) is used for the radar rainfall classes. If the coldest collocated IR pixel within 25 km of the radar location for the image before, during, and after the radar rainfall estimate is not colder than this threshold, then the estimate is regarded as false. Results for screening radar rainfall cases in this way have been very encouraging. Moderate and heavy radar rainfall cases are screened very well in virtually all cases. Light radar rainfall cases (0-5 mm hr-1 ) are the most difficult to screen.
4.0 Precipitation Analysis Products and Applications
The near-real-time daily precipitation analysis has been used to develop a number of additional products and applications. Some of these are described in the following subsections.
Precipitation analysis products developed at CPC include a daily precipitation analysis and associated station map; precipitation monitoring maps that highlight hydrologic anomalies over the conterminous U.S. for the previous 30 days and 90 days (Fig. 7), a series of products that verify precipitation forecasts from the operational MRF and ensembles, and a drought/flood potential product that highlights expected changes in observed precipitation anomalies. All of these products are disseminated on a daily basis via the CPC Web Site (http://www.cpc.ncep.noaa.gov/products/precip/realtime/). These products are undergoing continuous development and improvement and have benefitted significantly from the extended QC initiatives described in section 3.
The near-real-time precipitation analysis is used by several other CPC projects, including:
(1) The U.S. National Threats Assessment http://www.cpc.ncep.noaa.gov/products/expert_assessment/threats.html
(2) The U.S. Drought Assessment http://www.cpc.ncep.noaa.gov/products/expert_assessment/drought_assessment.html
(3) The Palmer Drought Index http://www.cpc.ncep.noaa.gov/products/monitoring_and_data/drought.html
(4) The CPC Soil Moisture Forecast project http://www.cpc.ncep.noaa.gov/soilmst/forecasts.html
The analyses are also used by external projects, including:
(1) The Land Data Assimilation System (LDAS)
http://ldas.gsfc.nasa.gov/index.shtml
(2) The NCEP Regional Reanalysis
http://wesley.ncep.noaa.gov/reanalysis.html
In the case of LDAS, an early analysis (based only on RFC data) is provided on a daily basis.
As a result of these collaborations, we are frequently required to adapt the content, design and availability of our precipitation products as well as to respond to changing user requirements.
4.3 Diagnostic Studies of the North American Monsoon System
Underpinning the effort discussed in sections 2 and 3 is a program of applied research to improve forecasts and real time monitoring of the North American monsoon. Major activities include a PACS-funded study of the annual cycle and seasonal-to-interannual variability of the North American Monsoon System (Higgins et al. 1997; Higgins et al. 1998; Higgins et al. 1999; Higgins and Shi 2000a; 2000b); a PACS-GCIP funded project to develop a near-real-time monitoring capability over the Americas; and the North American Monsoon Experiment (NAME) (Higgins et al. 2000). Details about each of these projects and links to the monsoon monitoring tools are found on the CPC Web site at the URL: http://www.cpc.ncep.noaa.gov/products/precip/monsoon/index.html
4.4 Precipitation Analyses for the Americas
Under the auspices of the CLIVAR/PACS and GCIP programs, the CPC has undertaken a comprehensive program to improve the analysis of gauge-based precipitation over the Americas on a range of space and time scales. The goal is to develop near-real-time and historical precipitation analyses for all of the Americas. The approach has been incremental, by first focusing on the U.S. and then by expanding this effort to include the remainder of North, Central and South America. Several gridded daily analysis products are currently available:
Near-Real Time Analyses Historical Reanalysis
United States (1996-present) United States (1948-1998; daily)
South America (1999-present) United States (1948-present; hourly)
Mexico (1961-1999)
Brazil (1961-1999)
Canada (1961-1999)
Additional information is available at http://www.cpc.ncep.noaa.gov/products/precip/realtime/. Each of these datasets will be updated, maintained, archived and distributed throughout the life time of NAME.
A Unified Raingauge Dataset (URD) has been developed from multiple sources of U.S. raingauge data. The URD has been used to produce daily, monthly and seasonal retrospective analyses for the period 1948-1997 (see section 6).
Three major sources of U.S. raingauge data are the basis of the URD. The first of these is the CPC Cooperative Dataset (consists of the first order" WMO GTS sites and the SHEF-encoded precipitation reports received via AFOS from the River Forecast Centers). Currently there are about 7000 active daily raingauge reports, with all but about 500 of these received from the River Forecast Centers. The period of record is February 1992-present. This dataset is CPC's major source of gauge data for the near-real-time precipitation analysis (section 3). Roughly 4% of the in situ reports in this dataset are delayed by more than 1 day, and hence are not used in our near-real-time daily analysis. The station network for the CPC Cooperative Dataset (for the period 1996-1999) is shown in Fig. 8b. The total number of stations that ever reported is 15,622. Most of these data are on a 1200Z-1200Z window, but in some cases the reporting time is ambiguous. About 250 stations do not have coordinates in the master list.
The second source of raingauge data is a compilation of daily observations by the National Climatic Data Center (NCDC, 1994) obtained from state universities, state cooperatives and the NWS (hereafter NCDC Cooperative Dataset). Currently there are about 8000 active daily raingauge reports, and the period of record is 1948-1998. The station network for the NCDC Cooperative Dataset is shown in Fig. 8a. The total number of stations that ever reported is 16,139 (after removing 805 stations with zero coordinates). The reporting time of the stations is sometimes ambiguous.
The third source of raingauge data was used to compile the multi-year (1948-present) gridded hourly precipitation dataset (hereafter HPD) of Higgins et al. 1996. This dataset was developed on hourly data obtained from the NWS Techniques Development Laboratory, who compiled, formatted and QC'ed data archived at the NOAA/NCDC. Currently there are about 2500 active hourly sites. The station network for the HPD dataset is shown in Fig. 8c. Typically, these data are available with a lag of ~1 year. The total number of stations that ever reported is 5933.
The URD has 3 versions. Version 1 consists of all gauge data (section 5.1) in a common format on a 1200Z-1200Z window. Hourly reports in the HPD dataset were aggregated into daily accumulations (1200Z-1200Z window) at the station level for insertion into the dataset. The HPD daily, NCDC daily and CPC daily gauge data were converted to an optimal monthly file format:
station ID, lat, lon, year, month, daily precipitation for the month
This format minimizes the size of Version 1, which is ~120 mb yr-1. The CPU workload to generate this version of the dataset was roughly 7 mb yr -1.
In Version 2 we eliminated duplicates from Version 1. Duplicate stations were defined as those with unique values but identical coordinates, or those with identical values and identical coordinates. A 1996 pilot study was carried out to study the general characterisitics of the overlap between datasets. Based on the pilot study, the general characteristics of the overlap were as follows:
21% of the CPC stations were in the NCDC list;19% of the NCDC stations were in CPC list
9% of the CPC stations were in the HPD list; 23% of the HPD stations were in the CPC list
30% of the NCDC stations were in the HPD list; 82% of the HPD stations were in the NCDC list
0.5% of the stations were in all three lists.
Thus, there was a significant amount of independent data in each source (i.e. further justification for Versions 2 and 3 of the URD). On a typical day the URD has ~13000-15000 stations. Thus the URD increases the number of daily reports available for retrospective analysis by roughly a factor of three (if compared to existing datasets used for historical and real-time analyses).
Based on the pilot study an algorithm was developed to eliminate duplicates for Version 2. Basically, once a duplicate was identified, the decision was based on whether the observation time was known accurately. This gave higher priority to the HPD data over the NCDC and CPC data, since the HPD observation time was always known. When NCDC and CPC duplicates were found, we usually accepted the CPC data because in most cases the NCDC station histories were well established. We note that the annual average precipitation difference between an analysis in which the NCDC data was accepted over the CPC data - versus one in which the CPC data was accepted over the NCDC data was quite small (the magnitude was less than 0.1 mm day-1 everywhere), indicating that the impact of these decisions on the overall dataset is minimal.
Version 3 is a gridded (0.25ox0.25o) analysis of Version 2 (see section 6). QC algorithms from the operational "near-real-time" analysis system were used to the extent possible. QC flags were inserted back into Version 2.
In the near future we will incorporate two additional datasets into the URD. The first of these is the Hydrometeorological Automated Data System (HADS) dataset, which consists of hourly data for the period 1992-present. The second is the SNOTEL dataset, which consists of gauge sites in the intermountain west. These are daily data for the period 1979-present. Other datasets will be incorporated as they become available. All data sets will be updated.
Since the autumn of 1999 we have been collaborating with the Office of Hydrology (J. Schaake), NCEP/EMC (K. Mitchell, Y. Lin) and members of the LDAS Project (K. Mitchell, D. Lohmann) on orographic precipitation adjustment in the western U.S. Currently we are testing an OH scheme (inverse distance weighting plus PRIZM) against the Cressman scheme. One possible outcome of these intercomparisons might be another precipitation reanalysis (see section 6.0) at higher resolution with an improved analysis scheme.
6.0 Daily Precipitation Reanalysis
Hydrologic anomalies cross monthly boundaries and occur at submonthly timescales. Thus, it is necessary to have a near-real-time daily precipitation analysis that is "well-connected" to a historical precipitation analysis, in order to place current anomalies in the proper historical context. Towards this end, we have completed a multi-year (1948-1997) daily precipitation reanalysis for the U.S. using the URD; the reanalysis will be extended forward as we are able to fill in missing data in the NCDC Cooperative Dataset. Our QC initiatives have been applied to the extent possible in generating the precipitation reanalysis. All QC flags were inserted back into Version 2 of the URD.
In the following subsections we present some statistics that illustrate the quality of the reanalysis product at various temporal scales. The reanalysis is compared to an existing climatology (based on the Hourly Precipitation Data) that is currently in use at CPC. For convenience, in the following subsections the two analyses will be referred to as the URD and the HPD.
The characteristics of the analysis scheme used for the precipitation reanalysis are identical to those in our operational near-real-time analysis (section 2.2). The daily analyses are gridded at a horizontal resolution of (lat,lon) = (0.25ox0.25o) over the domain 140o W - 60o W, 20o N - 60o N using a Cressman (1959) scheme with modifications (Glahn et al. 1985; Charba et al. 1992). The analysis on Day 1 is valid for the 24-hour window from 1200Z on day 0 to 1200Z on day 1.
In an annual mean sense the wettest parts of the country are the Pacific Northwest and the central Gulf Coast states (Figs. 9a, 9b). The structure of the rainfall pattern in the western U.S. is more realistic in the URD than in the HPD, where orographic precipitation and rain-shadow effects are resolved. This is clearly evident in the difference map (Fig. 9c). Note that the URD precipitation tends to be larger at various locations in the U.S., except at rain shadow locations in the west.
6.3 Monthly and Seasonal Variability
The seasonal cycle of monthly precipitation over the U.S. (Figs. 10 and 11) shows that the centers of maximum rainfall are in the Northwest and the Southeast during the cool season, and in the Southeast during the warm season. During the cool season the regions of maximum rainfall are strongly afected by winter cyclones. During spring the Pacific Northwest rainfall decreases but the Gulf Coast region continues to have high amounts due to the increasing role of convective activity. Summer precipitation in the west is very light. Throughout the year the lowest amounts occur in the Rockies, the High Plains, the Great Basin and the Desert Southwest. The largest differences between the URD and the HPD are in the western United States during the cold season, where the URD resolves heavy orographic precipitation, and in the southeastern U.S. during the warm season, where the additional stations in the URD bring up the analysis (Fig. 12). In all months of the year the URD analysis gives larger precipitation amounts than the HPD, except in the rain shadow regions of the western U.S. where the URD gives less precipitation.
Comparisons of 5-day running mean precipitation for selected regions in the U.S. (Fig. 13) show considerable differences among these regions in URD and HPD (Figs. 14 (URD), 15(HPD) and 16 (URD-HPD)); grid boxes over water are not included in the averages. For instance, the rainfall distribution of the Pacific Northwest is out of phase with that in the Midwest and the Great Plains states, while the Rocky Mountain and Northeast regions have little seasonal variation. Although the Pacific Northwest has a pronounced dry season, that region experiences the highest mean annual rainfall because the most intense precipitation rates (more than 5 mm day-1) are observed there during the cold season. The five-day running mean precipitation difference between the URD and the HPD (Fig. 16) again shows that the URD produces more precipitation throughout the annual cycle in each region. The mean difference between the URD and the HPD in each region shows that the largest increase is in the Pacific Northwest while the smallest increase is in the Inter-Mountain west.
The pattern of mean annual frequency of daily precipitation in the URD (Fig. 17a) is similar in many respects to the pattern of mean annual precipitation amount. The frequency of occurrence is highest in the Pacific Northwest despite the pronounced dry season there during the summer months. The frequency of occurrence is less than 15% in the Southwest, and exceeds 25% from the eastern Great Plains to the East Coast and in the Pacific Northwest.
The mean annual frequency of daily precipitation is lower in the URD than in the HPD (Fig. 17c) over most of the U.S. Precipitation events in the URD are more discrete (due to higher resolution and large input data base) whereas they tend to be smeared in the HPD (due to coarser resolution and less input data).
Next we examine the seasonal frequency of daily precipitation for various threshold precipitation rates. The thresholds are defined in terms of daily mean rates (i.e. mm day-1) and frequencies are expressed as a percentage of the total number of days in a season. In winter (JFM) the frequency of measurable rainfall (> 1 mm day-1) in the URD shows the highest values (60% or more) in the Pacific Northwest where Pacific cyclones come inland (Fig. 18). There are also relatively large values downwind of the Great Lakes (e.g. north central New York) where lake induced snow often persists for days after a winter storm and during cold air outbreaks. In summer (JAS) the frequency field has its highest values (60% or more) over Florida, consistent with the prevalence of showers and thunderstorms there.
As was the case for the annual mean, the mean seasonal frequency of daily precipitation is generally lower in the URD than in the HPD (Fig. 19). There are regional exceptions to this however, which highlight some of the improvements in the URD. Frequencies are higher in the URD in the Great Lakes in autumn and winter, where lake effect snows are resolved. They are also higher in the vicinity of high topography in the west (autumn, winter and spring) and in the southwest (summer) where orographic enhancement is resolved. They are decreased considerably in the traditional rain shadow regions of the west.
Peak frequencies for daily rainfall exceeding 10 mm day-1 in the URD (Fig. 20) are observed in the Pacific Northwest (more than 30%) during winter; maxima in this region are also found during autumn and spring. A secondary maximum is observed in the Southeast throughout the annual cycle, though frequencies are largest for the central Gulf Coast states during winter and for Florida during summer. The summer pattern shows the highest frequencies along the Gulf Coast from southern Louisiana to Florida, with another maximum over the upper midwest due mainly to mesoscale convective systems.
Frequencies for daily rainfall exceeding 10 mm day-1 are generally lower in the URD than in the HPD (Fig. 21), especially in the winter and spring. Frequencies are higher along the immediate Pacific Northwest coast (autumn, winter and spring), on the windward side of high topography in the Pacific Northwest (autumn, winter and spring), along the southeast coast (summer) and in the monsoon region of the Southwest (summer).
For daily rainfall rates exceeding 25 mm day-1(~1.0 inch day-1) in the URD (Fig. 22), the peak frequencies are found along the Pacific Northwest Coast and over interior mountain ranges of the Pacific Northwest and northern California during winter. A similar maximum also occurs along the Pacific Northwest coast in autumn. In spring the peak frequencies are along the central Gulf Coast states and western Tennessee Valley. In the autumn this maximum is shifted somewhat further to the east. There is a peak in heavy rain events during summer over Florida and along the immediate Gulf Coast and East Coast. Much os the western U.S. has frequencies of less than 1% throughout the annual cycle.
Frequencies for daily rainfall exceeding 25 mm day-1 are generally lower in the URD than in the HPD (Fig. 23), especially in the winter and spring. Frequencies are higher along the immediate Pacific Northwest coast (autumn and winter), on the windward side of high topography in the Pacific Northwest (autumn and winter), and along the southeast coast (summer).
Here we examine the standard deviation of daily precipitation in each season. Since the standard deviation is affected by the underlying march of the annual cycle, the daily climatology based on 1948-1997 has been removed from the daily data prior to the analysis.
The standard deviation of daily mean precipitation in the URD (Fig. 24) shows a very different pattern for summer as compared to the other seasons. In autumn, winter and spring the highest values are found in the Pacific Northwest due to Pacific storms, and in the central Gulf Coast States and lower Mississippi Valley due to synoptic-scale "Gulf" systems. In summer the variability in these regions is smaller (especially the Pacific Northwest). The largest variability in summer is in the central Great Plains and Midwest, where mesoscale convective systems (especially at night) can produce very heavy rains, and along the eastern seaboard.
A comparison of the standard deviation of daily mean precipitation in the URD and HPD (Fig. 25) shows that in general the variability of daily precipitation is larger in the URD (~ 2 mm day-1) over the eastern half of the U.S. The URD resolves topographic features in the western U.S. (e.g. rainshadows) not captured in the HPD.
The daily precipitation reanalysis, generated using Version 2 of the URD, is currently available. However, the size of the dataset (1.4GB compressed) is prohibitive. While CPC cannot assume the responsibility of servicing the data needs of the climate community, we have made arrangements with several established data distribution centers. CDC has transferred the dataset to NetCDF format and is currently hosting it on their web site (http://www.cdc.noaa.gov/cdc/data.unified.html). UCAR JOSS is hosting the dataset under the CODIAC system (http://www.joss.ucar.edu/codiac). CPC has also contacted NCDC about hosting the dataset.
A large portion of the work described here was done in close cooperation and active collaboration with Ken Mitchell of the NCEP/EMC, John Schaake of the Office of Hydrology, Jess Charba of Techniques Development Laboratory, and in consultation with many other colleagues in the NCEP Climate Prediction and Environmental Modeling Centers. Thanks are extended to Cathy Smith of CDC and Steve Williams of UCAR/JOSS for their efforts in hosting the daily precipitation reanalysis. This work was partially supported by the NOAA Office of Global Programs under the GCIP Project (Grant 8R1DA118).
Ahnert, P., W. Krajewski, and E. Johnson, 1986: Kalman filter estimation of radar-rainfall field bias. Preprints 23rd Conference on Radar Meteorology, Snowmass, Co., Amer. Meteor. Soc., JP33-JP37.
Barnes, S. L., 1964: A technique for maximizing details in numerical weather map analysis. J. App. Met., 3, 396-409.
Charba, J. P., A. W. Harrell III and A. C. Lackner III, 1992: A monthly precipitation amount climatology derived from published atlas maps: Development of a digital data base. TDL Office Note 92-7, NOAA, U.S. Department of Commerce, 20 pp.
Cressman, G. P., 1959: An operational objective analysis system. Mon. Wea. Rev., 87, 367-374.
Fulton, R. A., J. P. Breidenbach, D.-J. Seo, and D. A. Miller, 1997: The WSR-88D rainfall algorithm. Wea. Forecasting
Gandin, L. S., 1963: Objective analysis of meteorological fields. Gidrometeor. Isdat., Leningrad. [Israel Program for Scientific Translations, Jerusalem, 1965, 242 pp.]
Glahn, H. R., T. L. Chambers, W. S. Richardson, and H. P. Perrotti, 1985: Objective map analysis for the local AFOS MOS Program. NOAA Technical Memorandum NWS TDL 75, NOAA, U. S. Department of Commerce, 34 pp.
Higgins, R. W., J. E. Janowiak and Y. Yao, 1996: A gridded hourly precipitation data base for the United States (1963-1993). NCEP/Climate Prediction Center ATLAS No. 1., 47 pp.
-------------, Y. Yao and J. Wang 1997: Influence of the North American Monsoon system on the United States summer precipitation regime. J.Climate, 10, 2600-2622.
-------------, K. C. Mo and Y. Yao, 1998: Interannual variability of the United States summer precipitation regime with emphasis on the southwestern monsoon. J. Climate, 11, 2582-2606.
-------------, Y. Chen and A. V. Douglas, 1999: Interannual variability of the North American warm season precipitation regime. J. Climate, 12, 653-680.
-------------, and W. Shi, 2000a: Dominant factors responsible for interannual variability of the Southwest Monsoon. J. Climate, 13, 759-776.
-------------, and W. Shi, 2000b: Intercomparison of the principal modes of intraseasonal and interannual variability of the North American monsoon system. J. Climate (in press).
-------------, R. W. and the NAME Project Science Team, 2000: A Science and Implementation Plan for a North American Monsoon Experiment (NAME). 72pp. [Available on the web at http://www.cpc.ncep.noaa.gov/products/precip/monsoon/NAME.html]
Joyce, R., P. Arkin and W. Higgins, 1998: Estimation of orographic precipitation over western North America using geostationary satellite observations. 9th Conference on Satellite Meteorology and Oceanography, 25-29 May, Paris, France. pp 61-64.
Kalnay, E., and co-authors, 1996: The NCEP/NCAR Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437-471.
National Climatic Data Center, 1994: Surface Land daily Cooperative. Summary of the Day (TD-3200). National Environmental Satellite and Data Information Service, NOAA, U.S. Department of Commerce
Office of Hydrology, 1992: Stage II precipitation processing. National Weather Service, NOAA, U.S. Department of Commerce, 16 pp.
Seo, D.-J., R. Fulton, J. Breidenbach, and E. Johnson, 1997: Real-time estimation of mean field bias in radar rainfall data -- a review of current techniques in the National Weather Service and a proposed alternative. J. Atmos. Ocean. Technol.
Shepard, D., 1968: A two-dimensional interpolation function for irregularly spaced data. ACM National Conference. 512-524.
Smith, J. A., M. L. Baeck, and M. Steiner, 1997: Final report: Hydrometeorological assessments of the NEXRAD rainfall algorithms. Submitted to Office of Hydrology, National Weather Service, NOAA, U.S. Department of Commerce.
Yarosh, E.S. W. Higgins and W. Shi, 2000: A Unified US daily precipitation dataset and its application for water budget studies. 25th Climate Diagnostics and Prediction Workshop, Palisades, New York (Oct. 2000) (4 pages)
Figure 1. Flow chart summarizing how U.S. raingauge data is processed at the Climate Prediction Center.
Figure 2. (a) Typical station distribution for daily reporting stations in the United States from the CPC Cooperative Dataset. The dataset consists of reports gathered by the River Forecast Centers (~6000-7000 daily reports) and the Climate Anomaly Data Base (~400-500 daily reports). (b) Daily precipitation analysis (Units: mm) based on 24-hr accumulations for the period from 1200Z December 13, 2000 - 1200Z December 14, 2000. The analysis is gridded at a horizontal resolution of 0.25 degrees.
Figure 3. Number of stations in the CPC Cooperative Dataset reporting no precipitation, less than two inches, less than four inches, and in excess of 10 inches of precipitation for the period January-March 1998.
Figure 4. Typical examples of bias in radar data (Units: mm). For these examples the hourly biased data were used to produce 24-hr accumulations on a grid with a horizontal resolution of 4-km. The 24-hr accumulations are valid for (a) 1200Z May 2, 1998 - 1200Z May 3, 1998 and (b) 1200Z May 8, 1998 - 1200 Z May 9, 1998.
Figure 5. (a) Mean of the coldest satellite IR pixel (K) found within 0.25o of the 0.1o hourly radar precipitation estimates (units: mm hr-1). Abscissa is the hourly radar precipitation estimate, based on May 1999. (b) Standard deviation of (a).
Figure 6. (a) Distribution (%) of cases of coldest IR pixel (within 0.25o of the 0.1o hourly radar precipitation estimates) in standard deviations (sigma=22.2K) colder than the mean coldest pixel of 262.9 K for collocated zero radar rainfall. (b) Same as (a) except warmer than the mean coldest pixel. (c) Distribution of cases of coldest IR pixel in standard deviations (sigma=8.2K) colder than the mean coldest pixel of 208.2 K for collocated radar rainfall > 25 mm hr-1. Same as (c) except warmer than the mean coldest pixel.
Figure 7. Observed precipitation (upper left), departure from normal (upper right), percent of normal (lower left) and normal precipitation (lower right) for the 90-day period ending 30 June, 1998. Results are based on CPC's daily precipitation analysis which is produced by the U.S. Precipitation QC System and Analysis.
Figure 8. Total number of stations that ever reported in (a) the NCDC Cooperative Dataset, (b) the CPC Cooperative Dataset and (c) the HPD dataset. Typically, the number of daily reports is about 8000, 7000 and 2500 in the NCDC, CPC and HPD datasets.
Figure 9. Mean (1948-1997) annual precipitation (units: mm day-1) in (a) URD, (b) HPD and (c) the precipitation difference (URD-HPD).
Figure 10. Mean (1948-1997) monthly precipitation (units: mm day-1) for each month of the year in the URD.
Figure 11. Mean (1948-1997) monthly precipitation (units: mm day-1) for each month of the year in the HPD.
Figure 12. Mean (1948-1997) monthly precipitation difference (units: mm day-1) for each month of the year between the URD and the HPD.
Figure 13. Regions of the U.S. over which the averages are done in Figs. 14-16.
Figure 14. Five-day running mean (1948-1997) precipitation (units: mm day-1) in the URD for selected regions of the U.S.
Figure 15. Five-day running mean (1948-1997) precipitation (units: mm day-1) in the HPD for selected regions of the U.S..
Figure 16. Five-day running mean (1948-1997) precipitation difference (units: mm day-1) between URD and HPD for selected regions of the U.S.
Figure 17. Mean (1948-1997) annual frequency of daily precipitation expressed as a percentage of the total number of days in a year in (a) the URD, (b) the HPD and (c) the difference between the URD and the HPD.
Figure 18. Mean (1948-1997) seasonal frequency of daily precipitation in the URD for daily mean rates greater than 1 mm day-1 expressed as a percentage of the total number of days in JFM, AMJ, JAS and OND.
Figure 19. Mean (1948-1997) seasonal frequency of daily precipitation for daily mean rates greater than 1 mm day-1expressed as a percent difference between the URD and the HPD in JFM, AMJ, JAS and OND.
Figure 20. Mean (1948-1997) seasonal frequency of daily precipitation in the URD for daily mean rates greater than 10 mm day-1expressed as a percentage of the total number of days in JFM, AMJ, JAS and OND.
Figure 21. Mean (1948-1997) seasonal frequency of daily precipitation for daily mean rates greater than 10 mm day-1expressed as a percent difference between the URD and the HPD in JFM, AMJ, JAS and OND.
Figure 22. Mean (1948-1997) seasonal frequency of daily precipitation in the URD for daily mean rates greater than 25 mm day-1expressed as a percentage of the total number of days in JFM, AMJ, JAS and OND.
Figure 23. Mean (1948-1997) seasonal frequency of daily precipitation for daily mean rates greater than 25 mm day-1expressed as a percent difference between the URD and the HPD in JFM, AMJ, JAS and OND.
Figure 24. Standard deviation of daily mean (1948-1997) precipitation (units: mm day-1) in the URD for JFM, AMJ, JAS and OND.
Figure 25. Standard deviation of daily mean (1948-1997) precipitation (units: mm day-1) expressed as the difference between the URD and the HPD for JFM, AMJ, JAS and OND.
