• Adler, R. F., and et al. , 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 11471167, doi:10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Alemohammad, S. H., , McColl K. A. , , Konings A. G. , , Entekhabi D. , , and Stoffelen A. , 2015: Characterization of precipitation product errors across the United States using multiplicative triple collocation. Hydrol. Earth Syst. Sci., 19, 34893503, doi:10.5194/hess-19-3489-2015.

    • Search Google Scholar
    • Export Citation
  • Bolvin, D. T., , Adler R. F. , , Huffman G. J. , , Nelkin E. J. , , and Poutiainen J. P. , 2009: Comparison of GPCP monthly and daily precipitation estimates with high-latitude gauge observations. J. Appl. Meteor., 48, 18431857, doi:10.1175/2009JAMC2147.1.

    • Search Google Scholar
    • Export Citation
  • Chen, S., and et al. , 2013: Evaluation and uncertainty estimation of NOAA/NSSL Next-Generation National Mosaic Quantitative Precipitation Estimation Product (Q2) over the continental United States. J. Hydrometeor., 14, 13081322, doi:10.1175/JHM-D-12-0150.1.

    • Search Google Scholar
    • Export Citation
  • Dong, X., and et al. , 2011: Investigation the 2006 drought and 2007 flood extreme events at the SGP through an integrative analysis of observations. J. Geophys. Res., 116, D03204, doi:10.1029/2010JD014776.

    • Search Google Scholar
    • Export Citation
  • Fulton, R., , Breidenbach J. , , Seo D. J. , , Miller D. , , and O’Bannon T. , 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13, 377395, doi:10.1175/1520-0434(1998)013<0377:TWRA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Gebremichael, M., , Krajewski W. F. , , Morrissery M. L. , , Huffman G. J. , , and Adler R. F. , 2005: A detail evaluation of GPCP 1° daily rainfall estimates over the Mississippi River basin. J. Appl. Meteor., 44, 665681, doi:10.1175/JAM2233.1.

    • Search Google Scholar
    • Export Citation
  • Grams, H., 2013: NMQ update summer 2013: Important user information. Accessed 2 May 2016. [Available online at docs.google.com/document/d/1Op3uETOtd28YqZffgvEGoIj0qU6VU966iT_QNUOmqn4/edit.]

  • Huffman, G. J., , and Bolvin D. T. , 2013: Version 1.2 GPCP one-degree daily precipitation data set documentation. GPCP, 27 pp. [Available online at ftp://meso.gsfc.nasa.gov/pub/1dd-v1.2/1DD_v1.2_doc.pdf.]

  • Huffman, G. J., , Adler R. F. , , Morrissey M. , , Bolvin D. T. , , Curtis S. , , Joyce R. , , McGavock B. , , and Susskind J. , 2001: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 3650, doi:10.1175/1525-7541(2001)002<0036:GPAODD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and et al. , 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 3855, doi:10.1175/JHM560.1.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., , Adler R. F. , , Morrissey M. M. , , Bolvin D. T. , , and Gu G. , 2009: Improving the global precipitation record: GPCP Version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009GL040000.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., , Adler R. F. , , Bolvin D. T. , , and Nelkin E. , 2010: The TRMM Multi-Satellite Precipitation Analysis. Satellite Rainfall Applications for Surface Hydrology, F. Hossian and M. Gebremichael, Eds., Springer, 3–22.

  • Huffman, G. J., , Bolvin D. T. , , Braithwaite D. , , Hsu K. , , Joyce R. , , and Xie P. , 2014: NASA Global Precipitation Measurement (GPM) Integrated Multi-Satellite Retrievals for GPM (IMERG). NASA GSFC Algorithm Theoretical Basis Doc., version 4.4, 30 pp.

  • Joshi, M. K., , Rai A. , , and Pandey A. C. , 2013: Validation of TMPA and GPCP 1DD against the ground truth rain-gauge data for Indian region. Int. J. Climatol., 33, 26332648, doi:10.1002/joc.3612.

    • Search Google Scholar
    • Export Citation
  • Joyce, R. J., , Janowiak J. E. , , Arkin P. A. , , and Xie P. , 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487503, doi:10.1175/1525-7541(2004)005<0487:CAMTPG>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kummerow, C., , Olson W. S. , , and Giglio L. , 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 12131232, doi:10.1109/36.536538.

    • Search Google Scholar
    • Export Citation
  • Marshall, J. S., , Hitschfeld W. , , and Gunn K. L. S. , 1955: Advances in Radar Weather. Advances in Geophysics, Vol. 2, Academic Press, 1–56.

    • Search Google Scholar
    • Export Citation
  • McPhee, J., , and Margulis S. A. , 2005: Validation and error characterization of GPCP-1DD precipitation product over the contiguous United States. J. Hydrometeor., 6, 441459, doi:10.1175/JHM429.1.

    • Search Google Scholar
    • Export Citation
  • Rana, S., , McGregor J. , , and Renwick J. , 2015: Precipitation seasonality over the Indian subcontinent: An evaluation of gauge, reanalyses, and satellite retrievals. J. Hydrometeor., 16, 631651, doi:10.1175/JHM-D-14-0106.1.

    • Search Google Scholar
    • Export Citation
  • Rosenfeld, D., , Wolff D. B. , , and Atlas D. , 1993: General probability-matched relations between radar reflectivity and rain rate. J. Appl. Meteor., 32, 5072, doi:10.1175/1520-0450(1993)032<0050:GPMRBR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sorooshian, S., , Hsu K. L. , , Gao X. , , Gupta H. V. , , Imam B. , , and Braithwaite D. , 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 20352046, doi:10.1175/1520-0477(2000)081<2035:EOPSSE>2.3.CO;2.

    • Search Google Scholar
    • Export Citation
  • Stenz, R., , Dong X. , , Xi B. , , and Kuligowski R. J. , 2014: Assessment of SCaMPR and NEXRAD Q2 precipitation estimates using Oklahoma Mesonet observations. J. Hydrometeor., 15, 24842500, doi:10.1175/JHM-D-13-0199.1.

    • Search Google Scholar
    • Export Citation
  • Susskind, J., , and Pfaendtner J. , 1989: Impact of interactive physical retrievals on NWP. Report on the Joint ECMWF/EUMETSAT Workshop on the Use of Satellite Data in Operational Weather Prediction: 1989–1993, Vol. 1, T. Hollingsworth, Ed., ECMWF, 245–270.

  • Susskind, J., , Piraino P. , , Rokke L. , , Iredell T. , , and Mehta A. , 1997: Characteristics of the TOVS Pathfinder Path A dataset. Bull. Amer. Meteor. Soc., 78, 14491472, doi:10.1175/1520-0477(1997)078<1449:COTTPP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., , and Peters-Lidard C. D. , 2010: A global map of uncertainties in satellite-based precipitation measurements. J. Geophys. Res., 37, L24407, doi:10.1029/2010GL046008.

    • Search Google Scholar
    • Export Citation
  • Villarini, G., , Mandapaka P. V. , , Krajewski W. F. , , and Moore R. J. , 2008: Rainfall and sampling uncertainties: A rain gauge perspective. J. Geophys. Res., 113, D11102, doi:10.1029/2007JD009214.

    • Search Google Scholar
    • Export Citation
  • Wang, J., , Dong X. , , and Xi B. , 2015: Investigation of ice cloud microphysical properties of DCSs using aircraft in situ measurements during MC3E over the ARM SGP site. J. Geophys. Res. Atmos., 120, 35333552, doi:10.1002/2014JD022795.

    • Search Google Scholar
    • Export Citation
  • Wilson, J., , and Brandes E. A. , 1979: Radar measurement of rainfall—A summary. Bull. Amer. Meteor. Soc., 60, 10481058, doi:10.1175/1520-0477(1979)060<1048:RMORS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wu, W., , Kitzmiller D. , , and Wu S. , 2012: Evaluation of radar precipitation estimates from the National Mosaic and Multisensor Quantitative Precipitation Estimation System and the WSR-88D precipitation processing system over the conterminous United States. J. Hydrometeor., 13, 10801093, doi:10.1175/JHM-D-11-064.1.

    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 25392558, doi:10.1175/1520-0477(1997)078<2539:GPAYMA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Xie, P., , Janowiak J. E. , , Arkin P. A. , , Adler R. F. , , Gruber A. , , Ferraro R. , , Huffman G. J. , , and Curtis S. , 2003: GPCP pentad precipitation analyses: An experimental dataset based on gauge observations and satellite estimates. J. Climate, 16, 21972214, doi:10.1175/2769.1.

    • Search Google Scholar
    • Export Citation
  • Zhang, J., and et al. , 2011: National Mosaic and Multi-Sensor QPE (NMQ) system: Description, results, and future plans. Bull. Amer. Meteor. Soc., 92, 13211338, doi:10.1175/2011BAMS-D-11-00047.1.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (a) NMQ product domain (dotted dark blue box) divided into eight tiles. Dots of different colors represent different radar networks: WSR-88D (yellow and red), Terminal Doppler Weather Radar (TDWR; green), and the operational Canadian weather radars (light blue). Image adapted from Grams (2013). (b) Map of NEXRAD radar locations and coverage over United States. Image provided by the NOAA/NWS/Radar Operational Center.

  • View in gallery

    Each point represents a pair of collocated Q2 and GPCP 1DD daily precipitation estimates (excluding the samples for both datasets that equal 0.0 mm) during the period 2010–12 for each tile.

  • View in gallery

    As in Fig. 2, but for the warm season (April–September).

  • View in gallery

    As in Fig. 2, but for the cold season (October–March).

  • View in gallery

    As in Fig. 2, but for pairs of Q2 and GPCP 1DD monthly accumulated precipitation from all of their daily precipitation estimates during a month from 2010 to 2012.

  • View in gallery

    Each point represents a pair of Q2 and GPCP 1DD yearly accumulated precipitation estimates during the period 2010–12 for each tile.

  • View in gallery

    The histogram shows the distribution of GPCP 1DD (Q2) daily estimates when Q2 (GPCP 1DD) daily estimates are equal to 0.0 mm for each tile. In the legends, N0 represents the total number of daily estimates when both GPCP 1DD and Q2 equal to 0.0 mm; N1 and N2 represent the total numbers of daily estimates equal to 0.0 mm for GPCP 1DD and Q2, respectively. Also shown is the percentage of GPCP 1DD (Q2) estimates that are 0.0 mm given the corresponding Q2 (GPCP 1DD) estimate is equal to 0.0 mm.

  • View in gallery

    Spatially averaged precipitation from collocated Q2 and GPCP 1DD weighted daily estimates (excluding the samples for both datasets equal to 0.0 mm) in each tile during the period 2010–12. Each point represents the mean precipitation in each tile per day.

  • View in gallery

    Spatially averaged precipitation from Q2 and GPCP 1DD monthly accumulated precipitation in each tile during the period 2010–12. Each point represents the spatially averaged (over a tile) monthly accumulated precipitation per month.

  • View in gallery

    Average yearly precipitation from (a) Q2, (b) GPCP 1DD estimates, (c) their difference (GPCP − Q2), and (d) their normalized difference {(GPCP − Q2)/[(Q2 + GPCP)/2]} during the period 2010–12.

  • View in gallery

    Average warm season (April–September) precipitation from (a) Q2, (b) GPCP 1DD estimates, (c) their difference (GPCP − Q2), and (d) their normalized difference {(GPCP − Q2)/[(Q2 + GPCP)/2]} during the period 2010–12.

  • View in gallery

    Average cold season (October–March) precipitation from (a) Q2, (b) GPCP 1DD estimates, (c) their difference (GPCP − Q2), and (d) their normalized difference {(GPCP − Q2)/[(Q2 + GPCP)/2]} during the period 2010–12.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 153 153 10
PDF Downloads 54 54 3

Comparison of the GPCP 1DD Precipitation Product and NEXRAD Q2 Precipitation Estimates over the Continental United States

View More View Less
  • 1 Department of Atmospheric Sciences, University of North Dakota, Grand Forks, North Dakota
© Get Permissions
Full access

Abstract

This study compares the Global Precipitation Climatology Project (GPCP) 1 Degree Daily (1DD) precipitation estimates over the continental United States (CONUS) with National Mosaic and Multi-Sensor Quantitative Precipitation Estimation (NMQ) Next Generation (Q2) estimates. Spatial averages of monthly and yearly accumulated precipitation were computed based on daily estimates from six selected regions during the period 2010–12. Both Q2 and GPCP daily precipitation estimates show that precipitation amounts over southern regions (<40°N) are generally larger than northern regions (≥40°N). Correlation coefficients for daily estimates over selected regions range from 0.355 to 0.516 with mean differences (GPCP − Q2) varying from −0.86 to 0.99 mm. Better agreements are found in monthly estimates with the correlations varying from 0.635 to 0.787. For spatially averaged precipitation values averaged from grid boxes within selected regions, GPCP and Q2 estimates are well correlated, especially for monthly accumulated precipitation, with strong correlations ranging from 0.903 to 0.954. The comparisons between two datasets are also conducted for warm (April–September) and cold (October–March) seasons. During the warm season, GPCP estimates are 9.7% less than Q2 estimates, while during the cold season GPCP estimates exceed Q2 estimates by 6.9%. For precipitation over the CONUS, although annual means are close (978.54 for Q2 vs 941.79 mm for GPCP), Q2 estimates are much larger than GPCP over the central and southern United States and less than GPCP estimates in the northeastern United States. These results suggest that Q2 may have difficulties accurately estimating heavy rain and snow events, while GPCP may have an inability to capture some intense precipitation events, which warrants further investigation.

Corresponding author address: Professor Xiquan Dong, Department of Atmospheric Sciences, University of North Dakota, 4149 University Ave., Stop 9006, Grand Forks, ND 58202-9006. E-mail: dong@aero.und.edu

Abstract

This study compares the Global Precipitation Climatology Project (GPCP) 1 Degree Daily (1DD) precipitation estimates over the continental United States (CONUS) with National Mosaic and Multi-Sensor Quantitative Precipitation Estimation (NMQ) Next Generation (Q2) estimates. Spatial averages of monthly and yearly accumulated precipitation were computed based on daily estimates from six selected regions during the period 2010–12. Both Q2 and GPCP daily precipitation estimates show that precipitation amounts over southern regions (<40°N) are generally larger than northern regions (≥40°N). Correlation coefficients for daily estimates over selected regions range from 0.355 to 0.516 with mean differences (GPCP − Q2) varying from −0.86 to 0.99 mm. Better agreements are found in monthly estimates with the correlations varying from 0.635 to 0.787. For spatially averaged precipitation values averaged from grid boxes within selected regions, GPCP and Q2 estimates are well correlated, especially for monthly accumulated precipitation, with strong correlations ranging from 0.903 to 0.954. The comparisons between two datasets are also conducted for warm (April–September) and cold (October–March) seasons. During the warm season, GPCP estimates are 9.7% less than Q2 estimates, while during the cold season GPCP estimates exceed Q2 estimates by 6.9%. For precipitation over the CONUS, although annual means are close (978.54 for Q2 vs 941.79 mm for GPCP), Q2 estimates are much larger than GPCP over the central and southern United States and less than GPCP estimates in the northeastern United States. These results suggest that Q2 may have difficulties accurately estimating heavy rain and snow events, while GPCP may have an inability to capture some intense precipitation events, which warrants further investigation.

Corresponding author address: Professor Xiquan Dong, Department of Atmospheric Sciences, University of North Dakota, 4149 University Ave., Stop 9006, Grand Forks, ND 58202-9006. E-mail: dong@aero.und.edu

1. Introduction

Traditional ground-based precipitation measurements (e.g., rain gauges) directly observe precipitation and provide accurate rainfall estimates at point locations. However, in regions with sparse rain gauge coverage such as oceans and unpopulated areas, rain gauge measurements may not capture the spatial and temporal variability of precipitation events (Villarini et al. 2008). Radar provides high spatial and temporal resolution rainfall estimates, but several sources of uncertainty are associated with radar-based estimates, including attenuation, ground clutter, beam blockage, and variability in the reflectivity–rain rate (ZR) relationships (Wilson and Brandes 1979). Additionally, sufficient radar coverage for radar-based estimates is not widespread outside of the United States, Europe, and Japan. For complete global precipitation information, satellite-based estimates are relied upon; however, these often suffer from relatively large uncertainties (Xie and Arkin 1997; Huffman et al. 2001; Tian and Peters-Lidard 2010). Improvements in satellite instrument technology and precipitation estimation algorithms make it possible for researchers to continue to develop global precipitation products with finer temporal and spatial scales.

Satellite-based precipitation retrievals are typically calculated either from empirical relationships between measured cloud-top temperature from infrared (IR) instruments and precipitation rates or from passive microwave (PMW) instrument data that directly measure the scattering of upwelling radiation as well as the thermal emission from raindrops and hydrometers to estimate precipitation rate (Joyce et al. 2004). IR sensors on board geostationary satellites can provide precipitation estimates at high temporal resolution, but errors may occur where rainfall does not correlate well with cloud-top temperature (i.e., nonprecipitating cirrus clouds, tropical warm clouds). Microwave sensors measure precipitation in a more direct way but are only on board polar orbiters and therefore have a significant limitation in spatial and temporal sampling (Joyce et al. 2004). With the limitations intrinsic in the independent measurements in mind, new techniques have been developed to combine IR and PMW sensors on multiple satellites in order to produce global precipitation estimates by making use of the advantages of each sensor. Since the launch of Tropical Rainfall Measuring Mission (TRMM) in late 1997, precipitation techniques and algorithms have been devised and modified to combine IR and microwave observations to provide real-time precipitation information. To date, numerous satellite precipitation products have been released with various temporal and spatial resolutions, including Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN; Sorooshian et al. 2000), Climate Prediction Center morphing technique (CMORPH; Joyce et al. 2004), TRMM Multisatellite Precipitation Analysis (TMPA; Huffman et al. 2007, 2010), Integrated Multisatellite Retrievals for Global Precipitation Measurement (IMERG; Huffman et al. 2014]), and Global Precipitation Climatology Project (GPCP) 1 Degree Daily (1DD; Huffman et al. 2001). GPCP 1DD is based on the combination of data from multiple satellites and provides global daily precipitation estimates on a 1° grid box from 1996 to present. These long-term satellite-based precipitation estimates are critical in various scientific studies, including climate change analyses, hydrologic cycle studies, and validation of regional- and mesoscale numerical models (Huffman et al. 2001, 2009; Adler et al. 2003; Xie et al. 2003).

Adjustments and improvements have helped GPCP 1DD provide more reasonable precipitation estimates, such as the transition from TOVS data to AIRS data for precipitation estimates poleward of 40° latitude in April 2005 and the release of version 1.2 of the GPCP 1DD precipitation dataset in September 2012 (Huffman and Bolvin 2013). Nonetheless, the satellite-based techniques are associated with errors resulting from sampling data from different instruments and satellites, inaccurate estimates from precipitation algorithms under certain conditions, and the instruments themselves. Therefore, the accuracy of GPCP 1DD estimates should be determined through careful comparisons with high spatial and temporal resolution and coverage datasets. Several studies have been conducted to examine the performance of GPCP 1DD estimates over regions covered by dense ground-based networks (Huffman et al. 2001; McPhee and Margulis 2005; Gebremichael et al. 2005; Bolvin et al. 2009; Joshi et al. 2013). These results show that GPCP 1DD can reasonably represent the spatial and temporal rainfall distribution of the validation datasets. For more recent studies, Rana et al. (2015) evaluated GPCP 1DD (version 1.2), rain gauge, and reanalysis data against the Asian Precipitation–Highly Resolved Observational Data Integration Toward Evaluation of Water Resources (APHRODITE). They found that GPCP 1DD estimates show better performance than reanalysis data and are well correlated with APHRODITE. Alemohammad et al. (2015) investigated the uncertainties in stage IV Next Generation Weather Radar (NEXRAD-IV), TRMM 3B42 real time (RT), GPCP 1DD, and GPI precipitation products using triple collocation (TC) technique over the central part of the continental United States (CONUS). The results indicate that the GPCP 1DD, NEXRAD, and TRMM 3B42RT products show similar climatological patterns across the domain while GPI is different. However, GPCP 1DD and GPI products have poorer quality of precipitation estimates than NEXRAD and TRMM 3B42RT.

In this study, we compare the GPCP 1DD (version 1.2) precipitation product with the National Mosaic and Multi-Sensor Quantitative Precipitation Estimation (NMQ) Next Generation (Q2) over CONUS during the period 2010–12. Q2 composites radar data from the Weather Surveillance Radar-1988 Doppler (WSR-88D) network, producing instantaneous precipitation estimates at a high spatial and temporal resolution. Recent evaluation studies of the Q2 product suggest that Q2 estimates are viable as a validation tool for satellite precipitation retrievals in lieu of ground-based measurements in the future (Chen et al. 2013; Stenz et al. 2014). Uncertainties, however, still exist in radar-derived precipitation estimates, as mentioned in the beginning of this section. Since evaluations of Q2 estimates over large domains have only been performed against surface measurements (Wu et al. 2012; Chen et al. 2013), the comparison between radar-based Q2 estimates and satellite-based GPCP 1DD estimates not only directly shows the differences between the active and passive sensing data, but may also provide insight into the existing and potential limitations and strengths of each dataset. This study will use different time scales from daily to annual to compare the GPCP 1DD and Q2 estimates to investigate the possible errors that may be associated with Q2 and GPCP from short- to long-term estimates and show how differences between Q2 and GPCP estimates change with different time scales. Furthermore, utilizing the tiles created in the NMQ domain (Fig. 1a), the comparisons are conducted over six different regions for warm (April–September) and cold (October–March) seasons, which will help us gain a better understanding of the seasonal and spatial tendencies of each dataset and also determine which dataset might be more suitable over certain regions for certain precipitation types. It should be noted that this is not a validation study and neither GPCP 1DD nor Q2 estimates are treated as ground truth to validate the other. By comparing GPCP 1DD and Q2 estimates, this study will help researchers understand the uncertainties, regional characteristics, and strengths and limitations of each dataset, leading to better future use and the improvement of future versions of each dataset.

Fig. 1.
Fig. 1.

(a) NMQ product domain (dotted dark blue box) divided into eight tiles. Dots of different colors represent different radar networks: WSR-88D (yellow and red), Terminal Doppler Weather Radar (TDWR; green), and the operational Canadian weather radars (light blue). Image adapted from Grams (2013). (b) Map of NEXRAD radar locations and coverage over United States. Image provided by the NOAA/NWS/Radar Operational Center.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

This paper is organized into five sections. Section 2 gives a brief description of how the GPCP 1DD product and the Q2 product generate precipitation estimates and the comparison methodologies used in this study. Sections 3 and 4 present and discuss the results from multiple temporal (daily, monthly, and annual scale)- and spatial-scale comparisons. Finally, we present the conclusions in section 5.

2. Data and methodology

a. Datasets

GPCP 1DD, version 1.2, provides global daily precipitation estimates on a 1° × 1° grid from combined satellite observations. Between 40°S and 40°N, rainfall estimates are computed using the Threshold-Matched Precipitation Index (TMPI), which utilizes fractional occurrence of precipitation calculated by the Goddard profiling algorithm (GPROF; Kummerow et al. 1996), version 2004, using SSM/I–SSMIS data, the geostationary IR (geo-IR) brightness temperature histograms from the Geostationary Satellite Precipitation Data Centre (GSPDC), and precipitation rates computed by the GPI based on low-Earth orbiting IR (LEO-IR) data. The TMPI provides GPI-like precipitation estimates, but unlike the GPI, which assigns a conditional rain rate to all pixels within a given region that have a brightness temperature below a brightness temperature threshold , TMPI allows both and conditional rain rate to vary with space and time (Huffman et al. 2001). For each grid box, is obtained by accumulating the geo-IR histograms until the fraction of pixels matches the occurrence of precipitation. The conditional rain rate is then calculated by dividing GPCP satellite–gauge (SG) monthly estimates by the frequency of pixels below the . The rain-rate estimates from LEO-IR GPI are processed to fill in gaps where there is no geo-IR data. Poleward of 40°S and 40°N, the precipitation estimates through 2005 are computed from TOVS data. Since April 2005, AIRS data have been used in place of TOVS data. The TOVS–AIRS technique estimates precipitation using a multiple-regression relationship between collocated rain gauge measurements and various atmospheric parameters, including cloud-top pressure, fractional cloud cover, and relative humidity profiles (Susskind et al. 1997; Susskind and Pfaendtner 1989). Additionally, monthly sums of both TMPI and TOVS–AIRS daily estimates are computed and matched to the monthly SG precipitation estimates, which will reduce the inconsistency between the products with different temporal scales.

Q2 combines information from all ground-based radars in the NEXRAD network and provides precipitation estimates with a horizontal resolution of 1 km × 1 km and a temporal resolution of 5 min over the CONUS (Chen et al. 2013). Reflectivity data from multiple radars are mosaicked onto a common three-dimensional grid, then each vertical reflectivity column is objectively analyzed and prescribed a precipitation type (i.e., convective, stratiform, warm rain) if precipitation is present. The classification type determines which ZR relationship will be assigned to each grid cell to estimate rainfall. Four ZR relationships are used in association with the precipitation type in Q2 (Zhang et al. 2011): convective, (Fulton et al. 1998); stratiform, (Marshall et al. 1955); warm rain, (Rosenfeld et al. 1993); and snow at the surface, (Zhang et al. 2011).

Figure 1a shows that the NMQ domain is divided into eight tiles/regions. Tiles 1 and 5 as well as the western 5° longitude of tiles 2 and 6 have been excluded in this study because of significant beam blockages over the mountainous regions, as demonstrated in Fig. 1b. Tiles 2–4 have northern and southern boundaries at 60° and 40°N, while tiles 6–8 are bounded by 40° and 20°N. Western and eastern boundaries range from 110° to 90°W (in this study it is from 105° to 90°W), including 90°–80°W for tiles 3 and 7 and 80°–60°W for tiles 4 and 8.

b. Methodology

Q2 daily precipitation estimates from 2010 through 2012 are calculated by summing the hourly Q2 accumulated precipitation estimates each day. The 1 km × 1 km Q2 data are averaged into a 1° × 1° grid box using bilinear interpolation in order to match the spatial resolution of GPCP 1DD. Next, concurring Q2 and GPCP grid boxes with 0.0 mm accumulated precipitation are excluded. Furthermore, in order to investigate how the difference between the two datasets behaves in different time scales, monthly and annual accumulated precipitation are calculated by simply summing the daily estimates. We also investigate the occurrence of daily precipitation estimates for each dataset by determining the distribution of GPCP (Q2) when Q2 (GPCP) daily estimates are equal to 0.0 mm. For spatially averaged precipitation, the weighted precipitation amounts are calculated for each grid box based on the latitude of that grid box. Then spatially averaged precipitation amounts were computed for daily and monthly estimates by adding up weighted precipitation values of grid boxes within the tile and divided by the total number of grid boxes. Scatterplots were made between the Q2 and GPCP 1DD for daily, monthly, and annual estimates, as well as spatial averages along with their corresponding linear regression equations for six selected regions. The short- and long-term estimates of these two datasets and their differences can be examined and compared by different time-scale analyses. Note that the Q2 daily estimates over some regions are facing a poor radar coverage issue, and the estimates are certainly impacted by the radar coverage; however, the problem is not as severe as over tiles 1 and 5. Therefore, these daily estimates (if they have values) are used in the comparisons, and their uncertainties or errors are discussed in this study.

The comparisons over the six selected tiles can also provide a thorough insight into the regional differences between two datasets and show how terrain variability and different precipitation types affect these two estimates. GPCP 1DD estimates utilize two different algorithms, one for regions equatorward of 40° latitude and a second one for regions poleward of 40°. Therefore, the analysis of these selected regions (three tiles are above 40°N and three tiles are below 40°N) might reveal the influence of different algorithms on GPCP estimates. In addition, because precipitation mechanisms vary with season, especially in the CONUS, it is insightful to also perform a seasonal analysis. Therefore, comparisons between GPCP 1DD and Q2 estimates were conducted for both the warm season (April–September) and the cold season (October–March; Wu et al. 2012). Spatial distribution averages were also calculated to study how well the GPCP estimates agree with the Q2 estimates in the spatial pattern of precipitation.

3. Results

a. Comparisons of daily, monthly, and yearly accumulated precipitation estimates

Figures 24 are scatterplots of daily accumulated precipitation estimates for annual, warm season, and cold season in each tile, respectively, during the period 2010–12. Each point represents a pair of collocated Q2 and GPCP 1DD daily precipitation estimates (excluding the samples where both GPCP and Q2 equal 0.0 mm). Figure 2 shows that the daily precipitation over the southern tiles (6–8) are generally larger than those over northern tiles (2–4), with the most precipitation falling in tile 7 and the least falling in tile 2. For the comparisons between Q2 and GPCP precipitation estimates, the correlation coefficients are slightly higher in southern tiles than in northern tiles, with the highest correlation (0.516) in tile 6 and the lowest (0.355) in tile 3. Q2 estimates exceed GPCP estimates by 0.25 mm in tile 2, 0.50 mm in tile 7, and 0.86 mm in tile 6, while GPCP estimates exceed Q2 estimates by 0.21, 0.99, and 0.06 mm in tiles 3, 4, and 8, respectively.

Fig. 2.
Fig. 2.

Each point represents a pair of collocated Q2 and GPCP 1DD daily precipitation estimates (excluding the samples for both datasets that equal 0.0 mm) during the period 2010–12 for each tile.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Fig. 3.
Fig. 3.

As in Fig. 2, but for the warm season (April–September).

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Fig. 4.
Fig. 4.

As in Fig. 2, but for the cold season (October–March).

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

During the warm season, as shown in Fig. 3, daily mean precipitation for both Q2 and GPCP are larger than annual daily mean precipitation, particularly in tiles 2 and 3, but their correlation coefficients are slightly lower for all tiles except for tile 8, where correlation values increase from 0.475 to 0.486. Similar to Fig. 2, Fig. 3 shows that the largest negative and positive differences between GPCP and Q2 precipitation are −1.35 in tile 6 and 0.84 mm in tile 4. For tile 8, the difference between GPCP and Q2 estimates is only 0.2 mm during the warm season; however, GPCP estimates are 8.1% larger than Q2 estimates during the cold season (Fig. 4). Meanwhile, except for tile 8, slightly higher correlation coefficients are found between GPCP estimates and Q2 estimates during the cold season (Fig. 4) compared to the warm season (Fig. 3). During the cold season, the difference between two datasets becomes smaller in tile 6 (0.21 mm), but a large difference still exists in tile 4 with GPCP estimates larger than Q2 estimates by 1.15 mm. As shown in these three figures, the precipitation amounts of northern tiles (2–4) are less than their related southern tiles (6–8) for annual and warm season. Tile 4 has more precipitation than its related southern tile 8 for cold season. Q2 estimates have more samples with higher precipitation values compared to GPCP estimates. There are a few samples when GPCP estimates are around 10 mm while Q2 estimates exceed 100 mm.

Figure 5 is similar to Fig. 2 except for annual monthly accumulated precipitation from Q2 and GPCP 1DD daily estimates. Each dot represents a matched pair of Q2 and GPCP monthly accumulated precipitation. The distributions of Q2 and GPCP monthly accumulated precipitation and their differences over six tiles in Fig. 5 are very similar to their annual daily counterparts in Fig. 2 except with higher correlation coefficients. Tile 2 has the highest correlation values of 0.855, followed by tile 6, tile 8, and tile 7 with correlation values of 0.819, 0.787, and 0.767, respectively, as shown in Fig. 5.

Fig. 5.
Fig. 5.

As in Fig. 2, but for pairs of Q2 and GPCP 1DD monthly accumulated precipitation from all of their daily precipitation estimates during a month from 2010 to 2012.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Table 1 presents the mean values and standard deviations for warm and cold seasons’ monthly accumulated estimates of GPCP and Q2 and their correlation coefficients for each tile. For Q2 estimates, the standard deviations vary from 29.11 (tile 4) to 56.39 mm (tile 6) during the warm season. GPCP estimates have lower standard deviations for each tile, ranging from 20.56 (tile 3) to 40.77 mm (tile 8). During the cold season, lower standard deviations are found for both GPCP and Q2 estimates compare to the warm season, except for tile 4 of GPCP estimates (25.50 for warm season vs 31.75 mm for cold season). Strong correlations are still found in tile 6 for both the warm (0.760) and cold (0.858) seasons, and weaker correlations are found in tile 4 during the warm season (0.451) and in tile 3 during the cold season (0.418).

Table 1.

Mean values (mm) and standard deviations (mm) of monthly accumulated precipitation estimates of Q2 and GPCP 1DD and their correlations for each tile.

Table 1.

To further investigate the relationship between Q2 and GPCP estimates, yearly accumulated precipitation is displayed in Fig. 6. Similar to the monthly accumulated precipitation in Fig. 5, each point represents a pair of Q2 and GPCP yearly accumulated precipitation estimates. As presented in Fig. 6, the least amount of precipitation occurs in tile 2 for both Q2 and GPCP 1DD estimates (745.42 and 697.70 mm, respectively), while the greatest amount of precipitation is in tile 7 from Q2 estimates (1323.69 mm) and in tile 4 from GPCP estimates (1295.14 mm). The smallest difference (GPCP − Q2 = 13.55 mm) occurs in tile 8, whereas the largest one (233.87 mm) occurs in tile 4.

Fig. 6.
Fig. 6.

Each point represents a pair of Q2 and GPCP 1DD yearly accumulated precipitation estimates during the period 2010–12 for each tile.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Table 2 is similar to Table 1 except for yearly accumulated precipitation estimates. For the warm season, Q2 estimates vary from 560.54 (tile 6) to 770.59 mm (tile 7), and GPCP estimates have a range from 441.43 (tile 6) to 708.02 mm (tile 8). For the cold season, Q2 estimates range from 159.82 (tile 2) to 553.10 mm (tile 7), and GPCP estimates vary from 187.89 (tile 2) to 614.97 mm (tile 4). The highest standard deviations occur at tile 6 for both Q2 and GPCP estimates during warm (311.38 for Q2 vs 211.88 mm for GPCP) and cold (236.55 for Q2 vs 188.07 mm for GPCP) seasons. Consistent with its daily and monthly comparisons, tile 6 has the highest correlation coefficients (0.914, 0.837, and 0.925) for yearly, warm season, and cold season comparisons. Tile 4 has the lowest correlation coefficients for warm and cold seasons. Also for tile 4, the correlation is near zero between two datasets in terms of annual accumulated precipitations.

Table 2.

Mean values (mm) and standard deviations (mm) of yearly accumulated precipitation estimates of Q2 and GPCP 1DD and their correlations for each tile.

Table 2.

b. Comparison of daily precipitation occurrence

Figure 7 shows the probability distributions of GPCP (Q2) daily estimates where Q2 (GPCP) daily estimates are equal to 0.0 mm in six selected tiles. In Fig. 7, N0 represents the total number of daily estimates where both GPCP and Q2 equal to 0.0 mm, while N1 and N2 represent the total numbers of daily estimates equal to 0.0 mm for GPCP and Q2, respectively, during the 3-yr period. Also shown is the percentage of GPCP (Q2) estimates that are 0.0 mm given that the corresponding Q2 (GPCP) estimate is equal to 0.0 mm. For example, in tile 2, with a total of N2 = 99 231 samples for Q2 = 0.0 mm, 73.6% GPCP samples are equal to 0.0 mm and the other 26.4% GPCP samples are distributed from 0.0 to 15.0 mm. On the other hand, with a total of N1 = 94 009 for GPCP = 0.0 mm in tile 2, 77.7% Q2 samples are equal to 0.0 mm and the other 22.3% Q2 samples are distributed from 0.0 to 15.0 mm. The probability distributions for other tiles are similar to those in tile 2, ranging from 70% to 80% with a maximum co-occurrence of ~83% in tile 6. The differences between N1 and N2 for each tile are relatively small, ranging from 1147 (tile 8) to 5222 (tile 2).

Fig. 7.
Fig. 7.

The histogram shows the distribution of GPCP 1DD (Q2) daily estimates when Q2 (GPCP 1DD) daily estimates are equal to 0.0 mm for each tile. In the legends, N0 represents the total number of daily estimates when both GPCP 1DD and Q2 equal to 0.0 mm; N1 and N2 represent the total numbers of daily estimates equal to 0.0 mm for GPCP 1DD and Q2, respectively. Also shown is the percentage of GPCP 1DD (Q2) estimates that are 0.0 mm given the corresponding Q2 (GPCP 1DD) estimate is equal to 0.0 mm.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

c. Regional precipitation estimation comparisons

Figure 8 presents spatially averaged precipitation from collocated Q2 and GPCP 1DD daily weighted estimates in each tile during the period 2010–12. The mean precipitation in each tile is calculated from the sum of all daily weighted precipitation estimates (the weighted values are computed based on values used in Fig. 2, that is, excluding the samples where both Q2 and GPCP = 0.0 mm) within the tile, and then divided by the number of grid boxes. There are a total of 1096 points (2 × 365 + 366) in each tile during the 3-yr period. The mean values of Q2 and GPCP daily precipitation in Fig. 8 are different than those in Fig. 2 mainly because of the spatial weighting and spatial averaging. This also results in the slight changes in relative difference percentage between the two datasets. For example, in Fig. 2, GPCP estimates are larger than Q2 estimates by 22.2% for tile 4; meanwhile, in Fig. 8, this difference slightly increases to 25.1%. The significant difference between Figs. 2 and 8 is much higher correlation coefficients (0.638–0.788) and stronger linear relationships in Fig. 8, indicating that there are strong correlations for spatially averaged daily precipitation estimates. Figure 9 is similar to Fig. 8, but for spatially averaged weighted monthly accumulated precipitation for each tile. Similar to Fig. 8, the mean values in Fig. 9 are different from those in Fig. 5 because of the spatial weighting and, moreover, very strong correlations found between GPCP and Q2 estimates with a range of values from 0.903 in tile 3 to 0.954 in tile 2.

Fig. 8.
Fig. 8.

Spatially averaged precipitation from collocated Q2 and GPCP 1DD weighted daily estimates (excluding the samples for both datasets equal to 0.0 mm) in each tile during the period 2010–12. Each point represents the mean precipitation in each tile per day.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Fig. 9.
Fig. 9.

Spatially averaged precipitation from Q2 and GPCP 1DD monthly accumulated precipitation in each tile during the period 2010–12. Each point represents the spatially averaged (over a tile) monthly accumulated precipitation per month.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

To further investigate the precipitation distributions over the CONUS, the averages of yearly and seasonal accumulated precipitation as well as their difference (GPCP − Q2), and normalized difference percentage , during the period of 2010–12 are presented in Figs. 1012. For the yearly precipitation distribution, GPCP estimates have a similar pattern to Q2 estimates, with precipitation estimates increasing from west to east over the central United States and then decreasing slightly toward the eastern United States; however, Q2 estimates show more variation than GPCP estimates. Although the difference between their yearly precipitation estimates is only 36.75 mm (3.8%) over the CONUS, there are large regional differences between two datasets. Q2 estimates are much larger than GPCP estimates over the central United States, up to 600 mm larger in some areas, while in northern and northeastern regions Q2 estimates are less than GPCP estimates, as illustrated in Fig. 10c. GPCP estimates are 400 mm less than Q2 estimates over the Mississippi River basin, including Missouri, from the northeastern corner of Texas to northern Louisiana, northern Mississippi, and western West Virginia. The NDPs between GPCP and Q2 estimates are relatively small, with values between −10% and 10% for most of the study region.

Fig. 10.
Fig. 10.

Average yearly precipitation from (a) Q2, (b) GPCP 1DD estimates, (c) their difference (GPCP − Q2), and (d) their normalized difference {(GPCP − Q2)/[(Q2 + GPCP)/2]} during the period 2010–12.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Fig. 11.
Fig. 11.

Average warm season (April–September) precipitation from (a) Q2, (b) GPCP 1DD estimates, (c) their difference (GPCP − Q2), and (d) their normalized difference {(GPCP − Q2)/[(Q2 + GPCP)/2]} during the period 2010–12.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Fig. 12.
Fig. 12.

Average cold season (October–March) precipitation from (a) Q2, (b) GPCP 1DD estimates, (c) their difference (GPCP − Q2), and (d) their normalized difference {(GPCP − Q2)/[(Q2 + GPCP)/2]} during the period 2010–12.

Citation: Journal of Hydrometeorology 17, 6; 10.1175/JHM-D-15-0235.1

Figure 11 shows the distributions of the warm season accumulated precipitation from Q2 and GPCP, which resemble their yearly precipitation distributions where Q2 estimates vary from 0 to 1200 mm while GPCP estimates have a range of 0–1000 mm. GPCP estimates, on average, are 60.75 mm (NDP = −10.2%) less than Q2 estimates. For the cold season shown in Fig. 12, the spatial distribution pattern in Q2 estimates is almost the same as that in the GPCP estimates, with minor differences in some regions. Mean GPCP precipitation is 24.20 mm (NDP = 6.6%) larger than mean Q2 precipitation. For some areas over southern Montana, Wyoming, and southwestern Texas, where GPCP estimates are greater than Q2 estimates around 100 mm, the NDPs exceed −40%, as presented in Fig. 12d. From warm season to cold season, the precipitation amount estimated by Q2 decreases by 273.76 mm while GPCP decreases by 188.61 mm.

4. Discussion

The RDPs between Q2 and GPCP estimates remain consistent in each tile from daily to yearly comparisons (Figs. 2, 5, 6) for the period 2010–12. However, seasonal differences exist in tiles 2, 3, and 8 as the RDPs switch from negative in the warm season to positive in the cold season. The different algorithms of GPCP 1DD used to estimate the precipitation between northern and southern regions may contribute to the significant differences in RDPs, such as between tile 4 and its related tile, tile 8. Correlation coefficients for monthly analyses are much higher than those for daily analyses in all tiles, which is expected because of temporal averaging. However, when transitioning from monthly to yearly analyses in tiles 3, 4, 7, and 8, correlations actually decrease, which is counterintuitive. This may be because the differences between Q2 and GPCP monthly estimates are in different directions or highly variable.

From Fig. 3, we notice that during the warm season, GPCP estimates are significantly less than Q2 estimates in tile 2 (RDP = −12.83%) and tile 6 (RDP = −21.26%). Since tiles 2 and 6 encompass the plains region, much of the warm season rainfall is dominated by convective precipitation, particularly in tile 6. Because of the smaller scale of the convective systems and the fact that satellite instruments are limited in spatial resolution compared to radar observations, as well as the relatively sparse Global Precipitation Climatology Centre (GPCC) gauge coverage, GPCP estimates suffer from an inability to capture some daily intense precipitation events. Meanwhile, Q2 has a wet bias when estimating the accumulated precipitation from deep convective systems (DCSs) during the warm season (Stenz et al. 2014; Wang et al. 2015). Therefore, the actual value of precipitation during intense convection may fall between GPCP and Q2 estimates. As a result, the accumulated warm season precipitation estimated by GPCP is less than that from Q2 for daily to yearly accumulated precipitation (Fig. 3, Tables 1, 2) for all tiles except for tile 4, where DCSs are much less frequent than in the other tiles.

The Q2 and GPCP comparisons in tile 6 are consistent with the findings in Stenz et al. (2014) and Huffman et al. (2001). Q2 daily precipitation estimates were compared with Oklahoma Mesonet observations during the period 2010–12 with a correlation coefficient of 0.851 and a 5% wet bias during the cold season, which indicates that the Q2 precipitation estimates from NEXRAD reflectivity are reasonable for stratiform-dominated precipitation (Stenz et al. 2014). Meanwhile, GPCP estimates display a minimum in bias during winter when compared to Oklahoma Mesonet observations in Huffman et al. (2001), implying that stratiform precipitation characteristics during the cold season generally yield more accurate satellite-based estimates. Therefore, the higher correlations and smaller differences between GPCP and Q2 estimates during the cold season than the warm season in tiles 6 and 7 are anticipated because of the stratiform-dominated precipitation.

Most GPCP estimates are greater than Q2 estimates during the cold season, especially in the northern tiles (tiles 2–4) where the differences between GPCP and Q2 estimates are around 20%. Winter precipitation, particularly snowfall, is typically associated with shallow clouds that are close to the surface, making a radar beam more likely to overshoot falling precipitation, which may lead to underestimations in Q2 estimates. Furthermore, unlike typical warm season rainfall, which is classified into three separate groups, each with its own ZR relationship, there is only one ZR relationship for snowfall in Q2. This relationship is designed for orographic precipitation and stratiform frozen precipitation (Zhang et al. 2011) and is therefore likely not skillful at estimating the heavier convective snowfall events. Additionally, GPCP estimates tend to overestimate precipitation amounts during the cold season, which is likely because of the difficulty satellite instruments have when estimating frozen precipitation (McPhee and Margulis 2005; Bolvin et al. 2009).

In tiles 3 and 4, GPCP estimates are greater than Q2 estimates for the entire year. When compared to other tiles, a more significant portion of yearly precipitation occurs during the cold season in tiles 3 and 4 than in the remaining tiles. As presented in Table 2, the Q2 and GPCP precipitation estimates during the cold season contribute approximately 35% and 46% (Q2) and 41% and 47% (GPCP) to their yearly accumulated precipitation in tiles 3 and 4, respectively. Tiles 3 and 4 are located in the Great Lakes region and the northeastern United States, where frozen precipitation typically dominates during the winter. Thus, Q2 estimates likely underestimate the precipitation over these two tiles. For the warm season, mesoscale convective systems (MCSs) in tile 4 are less frequent and generally weaker than in tiles 6 and 7. As a result, MCSs may be misclassified as stratiform events, leading to the use of inappropriate ZR relationships and underestimation of precipitation amounts. For tile 2, although frozen precipitation is the dominant precipitation type during the cold season, a more significant portion of annual precipitation comes from warm season (79% for Q2, 73% for GPCP) than in tiles 3 and 4. Therefore, the difference between GPCP and Q2 precipitation estimates in tile 2 is much less than those in tiles 3 and 4.

In tile 8, the difference between GPCP and Q2 estimates is relatively small throughout the year. Tile 8 is located on the East Coast of the United States, next to the Atlantic Ocean. Abundant moisture from the ocean and tropical-like climate lead to less variability in precipitation type. The RDPs switch from positive to negative from warm to cold season; this is likely due to the snowfall events that occur at the northern part of the tile during the cold season, which leads to underestimates in Q2.

For spatially averaged comparisons, strong correlations exist between Q2 and GPCP estimates, which is expected because of large-scale averaging. The two datasets also show similar spatial patterns where the precipitation estimates increase from the western Great Plains to the central United States (95°–85°W), as illustrated in Figs. 8 and 9. The large precipitation estimate amount (≥1400 mm) over the central United States is primarily due to the strong moisture transport from the Gulf of Mexico by the low-level jet during the warm season (Dong et al. 2011). Because Q2 estimates are derived from radar retrievals and have a 1-km spatial and a 5-min temporal resolution, Q2 estimates are likely to capture more instantaneous precipitation events, particularly small-scale and short period events that are not detected by the satellite-derived GPCP estimates. Thus, Q2 estimates show more detail in precipitation variability. This can be seen on spatial distribution maps for Q2 estimates (Fig. 10a) and GPCP estimates (Fig. 10b). Most regions where Q2 estimates are significantly less than GPCP estimates (GPCP − Q2 > 400 mm shown in Fig. 10c) have poor or zero radar coverage and correspond to locations where the bottom of beam height is between ~1830 and ~3050 m or higher, as illustrated in Fig. 1b; some exceptions are regions in tile 4 where Q2 estimates are less than GPCP estimates, which is mainly due to the problems associated with ZR relationships and radar overshooting problem when estimating snowfall. For the regions where GPCP estimates are much larger than Q2 estimates, specifically from northern West Virginia down to northern Georgia, the Appalachian Mountains create beam blockage for local radars, and underestimation by Q2 likely occurs. During the cold season (Fig. 12d), the normalized differences between the two datasets over poor radar coverage regions are much greater than those during the warm season (Fig. 11d), which is more likely because of radar overshooting problems when shallow precipitation events occurs in regions of poor radar coverage during the cold season.

A 95% confidence interval is calculated for the mean differences between GPCP and Q2 estimates for each tile. Because of a lack of physical reasoning behind GPCP or Q2 being consistently different than the other dataset in one tile, a two-tailed test was chosen using a Z value of 1.96. The confidence interval of the difference is calculated by
e1
where is the mean difference of daily estimates between GPCP and Q2 (GPCP − Q2; mm), is the standard deviation of daily estimate difference, and is the total sample number of daily estimates for GPCP and Q2 in each tile. As shown in Table 3, the confidence intervals are relatively small because of the large sample size of daily estimates. Most of the confidence intervals have the same sign as their mean daily difference value, indicating that most of the GPCP estimates are either larger (if both bounds of interval are greater than 0.0 mm) or smaller (if both bounds of interval are less than 0.0 mm) than Q2 daily estimates. With the smallest sample size among tiles and least variation of sample variances, tile 8 shows the largest range of mean differences, and confidence intervals vary from negative to positive for year-round daily differences.
Table 3.

The mean differences between GPCP and Q2 daily estimates (GPCP − Q2; mm) and their 95% confidence interval (CI; mm) for each tile.

Table 3.

Compared to radar-based Q2 estimates, satellite-based GPCP estimates have coarse spatial and temporal resolutions. Furthermore, satellite instruments may have detectability issues with small-scale and short-duration precipitation events, especially during the warm season when intense convective precipitation events occur frequently. Therefore, Q2 estimates may provide more reasonable results than GPCP estimates over the plain regions during the warm season as long as the wet bias in Q2 estimates is taken into consideration. The excellent agreement between GPCP and Q2 cold season estimates for tiles 6 and 7 as well as strong correlations with Oklahoma Mesonet observations for both GPCP and Q2 estimates in previous studies indicate that GPCP and Q2 estimates can represent the stratiform rainfall during the cold season well. However, for ice phase precipitation—especially snow over northeastern regions—GPCP estimates may be more reliable than Q2 estimates. This could be because of the issues involved with radar overshooting and ZR relationship, resulting in the underestimation by Q2. Nevertheless, the accuracy of the TOVS–AIRS technique used in the GPCP 1DD product to estimate high-latitude cold season precipitation should be determined via comparison with the high-quality surface observations. Depending on the availability of radar coverage, Q2 estimates show a discontinuity in precipitation estimates, which is also noted in Stenz et al. (2014). When using Q2, the underestimation of precipitation amounts over regions with insufficient radar coverage should be taken into account. Fortunately, this is the kind of issue that can be avoided by satellite-based GPCP estimates.

5. Conclusions

This study compares the GPCP 1DD precipitation estimates with high-quality radar estimates from Q2 over the central and eastern CONUS covering a total of 619 grid boxes of 1° × 1°. The GPCP 1DD and Q2 precipitation estimates are compared over six different regions specified by the NMQ domain from January 2010 to December 2012. The temporal analyses include daily, monthly, and yearly accumulated precipitation estimates, and the spatial averages include annual, warm, and cold season estimates. In this study, we utilize the RDPs instead of NDPs to quantify the magnitude of differences between Q2 and GPCP estimates due to the relatively smaller values of NDPs that cannot represent the differences between these two datasets well. From the 3-yr comparisons between Q2 and GPCP precipitation estimates, we report the following conclusions.

  1. Both Q2 and GPCP 1DD daily precipitation estimates show that precipitation amounts over southern tiles (6–8) are generally larger than those over northern tiles (2–4), with the most precipitation in tile 7 and the least precipitation in tile 2. Monthly accumulated precipitation and differences in estimated precipitation over six selected tiles are very similar to their daily counterparts, except that correlation coefficients at the monthly scale (0.472–0.855) are much higher than those at the daily scale (0.355–0.516). However, this is not true moving from monthly to annual accumulated precipitation in tiles 4, 7, and 8 because the differences between Q2 and GPCP monthly estimates are in different directions and because of inadequate sample size of annual estimates over the relatively short study period. RDPs between Q2 and GPCP estimates remain consistent in each tile from daily to yearly comparisons. However, there are seasonal differences of RDPs that exist in tiles 2, 3, and 8 as they switch from negative in the warm season to positive in the cold season.
  2. During the warm season, the averages of daily precipitation from both Q2 and GPCP are generally larger than annual daily precipitation amounts, particularly in tiles 2 and 3, but correlation coefficients are slightly lower for all tiles except for tile 8. In contrast to the warm season comparisons, their counterparts during the cold season are much smaller than their annual and warm season daily precipitation estimates, especially in tile 2. Cold season correlation coefficients are slightly higher or close to their annual ones. During the warm season, the NDPs of GPCP estimates are 10.2% less than Q2 estimates, while during the cold season, the NDPs of GPCP estimates exceed Q2 estimates by 6.6%, resulting in −3.8% of the annual NDP for GPCP estimate over the CONUS.
  3. For spatially averaged precipitation in each tile, excellent agreements are found between GPCP and Q2 estimates, especially for monthly accumulated precipitation with strong correlations ranging from 0.903 to 0.954. Although the difference between yearly averaged precipitation estimates is only −36.75 mm (−3.76%) over the entire study region, there are large regional differences between GPCP and Q2 estimates. Q2 estimates are much larger than GPCP estimates over the central United States, while in northern and northeastern regions Q2 estimates are less than GPCP estimates.

Although there are several studies related to the evaluation of the GPCP 1DD product over the CONUS, this study focuses more on the comparison of GPCP 1DD and Q2 estimates and gains a better understanding of advantages, weaknesses, similarities, and differences between these widely used satellite- and radar-based precipitation estimates. In this study, for long-term estimates, GPCP 1DD and Q2 estimates both show fairly similar results; however, on seasonal and regional scales, large differences arise. For example, during the warm season in plains regions, Q2 estimates might be more reliable than GPCP estimates for convective precipitation events since they accurately capture the characteristics of DCSs (Stenz et al. 2014) as long as the wet bias of Q2 estimates is accounted for. For cold season precipitation in the northern United States, GPCP is likely a better choice for most applications because of radar overshooting issues and the use of one ZR relationship for all snowfall in Q2 estimates. Additionally, Q2 may largely underestimate the precipitation over regions with poor radar coverage, which is a problem the satellite-derived GPCP estimates avoid. These factors should be carefully considered when using these two products for different applications. We hope this study will provide valuable information to the strengths and weaknesses of each precipitation product and lead to better utilization by their users. Future work will be conducted on evaluating the long-term performance of precipitation estimates from recently developed reanalysis data over the CONUS using the GPCP 1DD product.

Acknowledgments

The NEXRAD Q2 product was obtained from the NOAA National Severe Storms Laboratory, and GPCP 1DD data were downloaded from the webpage http://precip.gsfc.nasa.gov/. This research was supported by NOAA Climate Program Office MAPP project with Award NA13OAR4310105 at the University of North Dakota.

REFERENCES

  • Adler, R. F., and et al. , 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 11471167, doi:10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Alemohammad, S. H., , McColl K. A. , , Konings A. G. , , Entekhabi D. , , and Stoffelen A. , 2015: Characterization of precipitation product errors across the United States using multiplicative triple collocation. Hydrol. Earth Syst. Sci., 19, 34893503, doi:10.5194/hess-19-3489-2015.

    • Search Google Scholar
    • Export Citation
  • Bolvin, D. T., , Adler R. F. , , Huffman G. J. , , Nelkin E. J. , , and Poutiainen J. P. , 2009: Comparison of GPCP monthly and daily precipitation estimates with high-latitude gauge observations. J. Appl. Meteor., 48, 18431857, doi:10.1175/2009JAMC2147.1.

    • Search Google Scholar
    • Export Citation
  • Chen, S., and et al. , 2013: Evaluation and uncertainty estimation of NOAA/NSSL Next-Generation National Mosaic Quantitative Precipitation Estimation Product (Q2) over the continental United States. J. Hydrometeor., 14, 13081322, doi:10.1175/JHM-D-12-0150.1.

    • Search Google Scholar
    • Export Citation
  • Dong, X., and et al. , 2011: Investigation the 2006 drought and 2007 flood extreme events at the SGP through an integrative analysis of observations. J. Geophys. Res., 116, D03204, doi:10.1029/2010JD014776.

    • Search Google Scholar
    • Export Citation
  • Fulton, R., , Breidenbach J. , , Seo D. J. , , Miller D. , , and O’Bannon T. , 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13, 377395, doi:10.1175/1520-0434(1998)013<0377:TWRA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Gebremichael, M., , Krajewski W. F. , , Morrissery M. L. , , Huffman G. J. , , and Adler R. F. , 2005: A detail evaluation of GPCP 1° daily rainfall estimates over the Mississippi River basin. J. Appl. Meteor., 44, 665681, doi:10.1175/JAM2233.1.

    • Search Google Scholar
    • Export Citation
  • Grams, H., 2013: NMQ update summer 2013: Important user information. Accessed 2 May 2016. [Available online at docs.google.com/document/d/1Op3uETOtd28YqZffgvEGoIj0qU6VU966iT_QNUOmqn4/edit.]

  • Huffman, G. J., , and Bolvin D. T. , 2013: Version 1.2 GPCP one-degree daily precipitation data set documentation. GPCP, 27 pp. [Available online at ftp://meso.gsfc.nasa.gov/pub/1dd-v1.2/1DD_v1.2_doc.pdf.]

  • Huffman, G. J., , Adler R. F. , , Morrissey M. , , Bolvin D. T. , , Curtis S. , , Joyce R. , , McGavock B. , , and Susskind J. , 2001: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 3650, doi:10.1175/1525-7541(2001)002<0036:GPAODD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and et al. , 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 3855, doi:10.1175/JHM560.1.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., , Adler R. F. , , Morrissey M. M. , , Bolvin D. T. , , and Gu G. , 2009: Improving the global precipitation record: GPCP Version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009GL040000.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., , Adler R. F. , , Bolvin D. T. , , and Nelkin E. , 2010: The TRMM Multi-Satellite Precipitation Analysis. Satellite Rainfall Applications for Surface Hydrology, F. Hossian and M. Gebremichael, Eds., Springer, 3–22.

  • Huffman, G. J., , Bolvin D. T. , , Braithwaite D. , , Hsu K. , , Joyce R. , , and Xie P. , 2014: NASA Global Precipitation Measurement (GPM) Integrated Multi-Satellite Retrievals for GPM (IMERG). NASA GSFC Algorithm Theoretical Basis Doc., version 4.4, 30 pp.

  • Joshi, M. K., , Rai A. , , and Pandey A. C. , 2013: Validation of TMPA and GPCP 1DD against the ground truth rain-gauge data for Indian region. Int. J. Climatol., 33, 26332648, doi:10.1002/joc.3612.

    • Search Google Scholar
    • Export Citation
  • Joyce, R. J., , Janowiak J. E. , , Arkin P. A. , , and Xie P. , 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487503, doi:10.1175/1525-7541(2004)005<0487:CAMTPG>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kummerow, C., , Olson W. S. , , and Giglio L. , 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 12131232, doi:10.1109/36.536538.

    • Search Google Scholar
    • Export Citation
  • Marshall, J. S., , Hitschfeld W. , , and Gunn K. L. S. , 1955: Advances in Radar Weather. Advances in Geophysics, Vol. 2, Academic Press, 1–56.

    • Search Google Scholar
    • Export Citation
  • McPhee, J., , and Margulis S. A. , 2005: Validation and error characterization of GPCP-1DD precipitation product over the contiguous United States. J. Hydrometeor., 6, 441459, doi:10.1175/JHM429.1.

    • Search Google Scholar
    • Export Citation
  • Rana, S., , McGregor J. , , and Renwick J. , 2015: Precipitation seasonality over the Indian subcontinent: An evaluation of gauge, reanalyses, and satellite retrievals. J. Hydrometeor., 16, 631651, doi:10.1175/JHM-D-14-0106.1.

    • Search Google Scholar
    • Export Citation
  • Rosenfeld, D., , Wolff D. B. , , and Atlas D. , 1993: General probability-matched relations between radar reflectivity and rain rate. J. Appl. Meteor., 32, 5072, doi:10.1175/1520-0450(1993)032<0050:GPMRBR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sorooshian, S., , Hsu K. L. , , Gao X. , , Gupta H. V. , , Imam B. , , and Braithwaite D. , 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 20352046, doi:10.1175/1520-0477(2000)081<2035:EOPSSE>2.3.CO;2.

    • Search Google Scholar
    • Export Citation
  • Stenz, R., , Dong X. , , Xi B. , , and Kuligowski R. J. , 2014: Assessment of SCaMPR and NEXRAD Q2 precipitation estimates using Oklahoma Mesonet observations. J. Hydrometeor., 15, 24842500, doi:10.1175/JHM-D-13-0199.1.

    • Search Google Scholar
    • Export Citation
  • Susskind, J., , and Pfaendtner J. , 1989: Impact of interactive physical retrievals on NWP. Report on the Joint ECMWF/EUMETSAT Workshop on the Use of Satellite Data in Operational Weather Prediction: 1989–1993, Vol. 1, T. Hollingsworth, Ed., ECMWF, 245–270.

  • Susskind, J., , Piraino P. , , Rokke L. , , Iredell T. , , and Mehta A. , 1997: Characteristics of the TOVS Pathfinder Path A dataset. Bull. Amer. Meteor. Soc., 78, 14491472, doi:10.1175/1520-0477(1997)078<1449:COTTPP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., , and Peters-Lidard C. D. , 2010: A global map of uncertainties in satellite-based precipitation measurements. J. Geophys. Res., 37, L24407, doi:10.1029/2010GL046008.

    • Search Google Scholar
    • Export Citation
  • Villarini, G., , Mandapaka P. V. , , Krajewski W. F. , , and Moore R. J. , 2008: Rainfall and sampling uncertainties: A rain gauge perspective. J. Geophys. Res., 113, D11102, doi:10.1029/2007JD009214.

    • Search Google Scholar
    • Export Citation
  • Wang, J., , Dong X. , , and Xi B. , 2015: Investigation of ice cloud microphysical properties of DCSs using aircraft in situ measurements during MC3E over the ARM SGP site. J. Geophys. Res. Atmos., 120, 35333552, doi:10.1002/2014JD022795.

    • Search Google Scholar
    • Export Citation
  • Wilson, J., , and Brandes E. A. , 1979: Radar measurement of rainfall—A summary. Bull. Amer. Meteor. Soc., 60, 10481058, doi:10.1175/1520-0477(1979)060<1048:RMORS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wu, W., , Kitzmiller D. , , and Wu S. , 2012: Evaluation of radar precipitation estimates from the National Mosaic and Multisensor Quantitative Precipitation Estimation System and the WSR-88D precipitation processing system over the conterminous United States. J. Hydrometeor., 13, 10801093, doi:10.1175/JHM-D-11-064.1.

    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 25392558, doi:10.1175/1520-0477(1997)078<2539:GPAYMA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Xie, P., , Janowiak J. E. , , Arkin P. A. , , Adler R. F. , , Gruber A. , , Ferraro R. , , Huffman G. J. , , and Curtis S. , 2003: GPCP pentad precipitation analyses: An experimental dataset based on gauge observations and satellite estimates. J. Climate, 16, 21972214, doi:10.1175/2769.1.

    • Search Google Scholar
    • Export Citation
  • Zhang, J., and et al. , 2011: National Mosaic and Multi-Sensor QPE (NMQ) system: Description, results, and future plans. Bull. Amer. Meteor. Soc., 92, 13211338, doi:10.1175/2011BAMS-D-11-00047.1.

    • Search Google Scholar
    • Export Citation
Save