• Adler, R. F., , A. J. Negri, , P. R. Keehn, , and I. M. Hakkarinen, 1993: Estimation of monthly rainfall over Japan and surrounding waters from a combination of low-orbit microwave and geosynchronous IR data. J. Appl. Meteor., 32, 335356.

    • Search Google Scholar
    • Export Citation
  • Adler, R. F., , G. J. Huffman, , and P. R. Keehn, 1994: Global tropical rain rate estimates from microwave-adjusted geosynchronous IR data. Remote Sens. Rev., 11, 125152.

    • Search Google Scholar
    • Export Citation
  • Adler, R. F., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 11471167.

    • Search Google Scholar
    • Export Citation
  • ESA, cited 2009: ESA/ESA GlobCover Project, led by MEDIAS-France. [Available online at http://ionia1.esrin.esa.int/index.asp.]

  • Ferraro, R., 1997: SSM/I derived global rainfall estimates for climatological applications. J. Geophys. Res., 102, 16 71516 735.

  • Ferraro, R., , N. C. Grody, , and J. A. Kogut, 1986: Classification of geophysical parameters using passive microwave satellite measurements. IEEE Trans. Geosci. Remote Sens., GE-24, 10081013.

    • Search Google Scholar
    • Export Citation
  • Ferraro, R., , E. A. Smith, , W. Berg, , and G. Huffman, 1998: A review of screening techniques for passive microwave precipitation retrieval algorithms. J. Atmos. Sci., 55, 15831600.

    • Search Google Scholar
    • Export Citation
  • Ferraro, R., and Coauthors, 2005: NOAA operational hydrological products derived from the AMSU. IEEE Trans. Geosci. Remote Sens., 43, 10361049.

    • Search Google Scholar
    • Export Citation
  • Gopalan, K., , N.-Y. Wang, , R. Ferraro, , and C. Liu, 2010: Status of the TRMM 2A12 land precipitation algorithm. J. Atmos. Oceanic Technol., 27, 13431354.

    • Search Google Scholar
    • Export Citation
  • Grody, N. C., 1991: Classification of snow cover and precipitation using the Special Sensor Microwave/Imager (SSM/I). J. Geophys. Res., 96, 74237435.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., and Coauthors, 2000: Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21, 13311364.

    • Search Google Scholar
    • Export Citation
  • Hou, A. Y., 2006: The Global Precipitation Measurement (GPM) mission: An overview. 2006 EUMETSAT Meteorological Satellite Conf, Helsinki, Finland, EUMETSAT, P.48. [Available online at http://gpm.gsfc.nasa.gov/science/sciencetalks/Hou_EUMETSAT06.pdf.]

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 520.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 2009: Improving the global precipitation record: GPCP version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009GL040000.

    • Search Google Scholar
    • Export Citation
  • Hulme, M., 1992: Rainfall changes in Africa: 1931–60 to 1961–90. Int. J. Climatol., 12, 685699.

  • Kummerow, C., and Coauthors, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 18011820.

    • Search Google Scholar
    • Export Citation
  • Liu, C., , E. Zipser, , D. Cesil, , S. Nesbitt, , and S. Sherwood, 2008: A cloud and precipitation feature database from 9 years of TRMM observations. J. Appl. Meteor. Climatol., 47, 27122728.

    • Search Google Scholar
    • Export Citation
  • McCollum, J., , and R. R. Ferraro, 2003: The next generation of NOAA/NESDIS SSM/I, TMI and AMSR-E microwave land rainfall algorithms. J. Geophys. Res., 108, 83828404.

    • Search Google Scholar
    • Export Citation
  • NASA, cited 2009a: NASA/MODIS Atmosphere Program. [Available online at http://modis-atmos.gsfc.nasa.gov/ECOSYSTEM/index.html.]

  • NASA, 2009b: NASA/Land Data Assimilation Systems. [Available online at http://ldas.gsfc.nasa.gov.]

  • Nesbitt, S. W., , E. J. Zipser, , and D. J. Cecil, 2000: A census of precipitation features in the tropics using TRMM: Radar, ice scattering, and lightning observations. J. Climate, 13, 40874106.

    • Search Google Scholar
    • Export Citation
  • New, M., , M. Todd, , M. Hulme, , and P. Jones, 2001: Review precipitation measurements and trends in the twentieth century. Int. J. Climatol., 21, 18991922.

    • Search Google Scholar
    • Export Citation
  • Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381394.

  • Romanov, P., , D. Tarpley, , G. Gutman, , and T. Carroll, 2003: Mapping and monitoring of the snow cover fraction over North America. J. Geophys. Res., 108, 8619, doi:10.1029/2002JD003142.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., , and C. D. Peters-Lidard, 2007: Systematic anomalies over inland water bodies in satellite-based precipitation estimates. Geophys. Res. Lett., 34, L14403, doi:10.1029/2007GL030787.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., , C. D. Peters-Lidard, , B. J. Choudhury, , and M. Garcia, 2007: Multitemporal analysis of TRMM-based satellite precipitation products for land data assimilation applications. J. Hydrometeor., 8, 11651183.

    • Search Google Scholar
    • Export Citation
  • Wang, N.-Y., , C. Liu, , R. Ferraro, , D. Wolff, , E. Zipser, , and C. Kummerow, 2009: TRMM 2A12 land precipitation product – Status and future plans. J. Meteor. Soc. Japan, 87, 237253.

    • Search Google Scholar
    • Export Citation
  • Wilheit, T. A., , C. Kummerow, , and R. Ferraro, 2003: Rainfall algorithms for AMSR-E. IEEE Trans. Geosci. Remote Sens., 41, 204214.

  • Xie, P., , and P. A. Arkin, 1998: Global monthly precipitation estimates from satellite-observed outgoing longwave radiation. J. Climate, 11, 137164.

    • Search Google Scholar
    • Export Citation
  • Zipser, E., , D. Cecil, , C. Liu, , S. Nesbitt, , and S. Yuter, 2006: Where are the most intense thunderstorms on Earth? Bull. Amer. Meteor. Soc., 87, 10571071.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Schematic of the general structure of the new generic unified screening methodology for microwave land precipitation retrieval algorithms. Three land surface types are presented; the scheme can adopt more.

  • View in gallery

    Aggregation methods of gridded information in the GPROFV6 with the (top) current and (bottom) prototype surface screening schemes.

  • View in gallery

    Regions with elevation ≥2 km (blue).

  • View in gallery

    The surface classes from the UMD 1-km Global Land Cover map to be used in GPROFV2 2A12 land precipitation retrievals with the prototype surface screening scheme: white, water; red, desert or semi/permanent snow/ice covered; yellow, semiarid; green, open grassland probably semiarid; and gray, other surface types.

  • View in gallery

    Comparison of using the (top) current, fixed-grid land/water mask and (bottom) prototype surface screening scheme via the UMD 1-km map and 1-km precision FOV mapping in GPROFV6 showing better inland water (blue) and coastal (yellow) regions (orbit 42496, at 1041:45–1214:08 UTC 30 Apr 2005).

  • View in gallery

    The GPROFV6 2A12 retrievals with the (top) current and (middle) prototype surface screening schemes and (bottom) PR 2A25 retrievals. Regions of no retrieval are in white (orbit 42496, 1041:45–1214:08 UTC 30 Apr 2005).

  • View in gallery

    The GPROFV6 2A12 rain flag maps with the (top) current and (bottom) prototype surface screening schemes. False snow detections (green) are removed in the prototype scheme. Also shown are rain (red) and no-rain land (pink) and water–oceanic regions (blue) (orbit 42496, 1041:45–1214:08 UTC 30 Apr 2005).

  • View in gallery

    Nonzero rain event detection comparisons between the current and proposed surface screening scheme applications in GPROFV6 2A12 in JJA and DJF. Values are (proposed GPROFV6 2A12 − current GPROFV6 2A12) from total rain events with rain >0 mm h−1.

  • View in gallery

    Comparisons between GPROFV6 2A12 and PR 2A25 in the regions where there are differences in the retrievals between the GPROFV6 2A12 with the current and proposed surface screening schemes in JJA and DJF. Values are (GPROFV6 2A12 − PR 2A25)/PR 2A25 from rain rates (mm day−1) averaged in each season.

  • View in gallery

    Comparisons between GPCPV2.1 and GPROFV6 2A12 with the (top) current and (bottom) prototype surface screening schemes in (left) JJA and (right) DJF. Values are (GPROFV6 2A12 − GPCPV2.1)/GPCPV2.1 from rain rates (mm day−1) averaged in each season.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 22 22 10
PDF Downloads 6 6 3

Prototyping a Generic, Unified Land Surface Classification and Screening Methodology for GPM-Era Microwave Land Precipitation Retrieval Algorithms

View More View Less
  • 1 Environmental Management Technology Research Group, Institut Teknologi Bandung, Bandung, Indonesia
  • 2 Earth System Science Interdisciplinary Center, University of Maryland, College Park, College Park, Maryland
  • 3 Earth System Science Interdisciplinary Center, University of Maryland, College Park, College Park, and NOAA/National Environmental Satellite, Data, and Information Service, Silver Spring, Maryland
© Get Permissions
Full access

Abstract

A prototype generic, unified land surface classification and screening methodology for Global Precipitation Measurement (GPM)-era microwave land precipitation retrieval algorithms by using ancillary datasets is developed. As an alternative to the current radiometer-determined approach, the new methodology is shown to be promising in improving rain detection by providing better surface-cover-type information. The early prototype new surface screening scheme was applied to the current version of the Goddard profiling algorithm that is used for the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (GPROFV6). It has shown improvements in surface-cover-type classification and hence better precipitation retrieval comparisons with TRMM precipitation radar level-2 (L2) (2A25) data and the Global Precipitation Climatology Project (GPCP) version-2.1 (GPCPV2.1) datasets. The new ancillary data approach removes the current dependency of the screening step on relatively different satellite-specific channels and ensures the comparability and continuity of satellite-based precipitation products from different platforms. This is particularly important for advancing the current state of precipitation retrieval over land and for use in merged rainfall products.

Corresponding author address: Arief Sudradjat, KK Teknologi Pengelolaan Lingkungan, Fakultas Teknik Sipil dan Lingkungan, Institut Teknologi Bandung, Jalan Ganesa 10, Bandung 40132, Indonesia. E-mail: ariefs@tl.itb.ac.id

Abstract

A prototype generic, unified land surface classification and screening methodology for Global Precipitation Measurement (GPM)-era microwave land precipitation retrieval algorithms by using ancillary datasets is developed. As an alternative to the current radiometer-determined approach, the new methodology is shown to be promising in improving rain detection by providing better surface-cover-type information. The early prototype new surface screening scheme was applied to the current version of the Goddard profiling algorithm that is used for the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (GPROFV6). It has shown improvements in surface-cover-type classification and hence better precipitation retrieval comparisons with TRMM precipitation radar level-2 (L2) (2A25) data and the Global Precipitation Climatology Project (GPCP) version-2.1 (GPCPV2.1) datasets. The new ancillary data approach removes the current dependency of the screening step on relatively different satellite-specific channels and ensures the comparability and continuity of satellite-based precipitation products from different platforms. This is particularly important for advancing the current state of precipitation retrieval over land and for use in merged rainfall products.

Corresponding author address: Arief Sudradjat, KK Teknologi Pengelolaan Lingkungan, Fakultas Teknik Sipil dan Lingkungan, Institut Teknologi Bandung, Jalan Ganesa 10, Bandung 40132, Indonesia. E-mail: ariefs@tl.itb.ac.id

1. Introduction

Precipitation is one of the most important fluxes in the global water and energy system. Not only is it in general the ultimate source of water for land hydrology, it also plays an important role in transporting the sun’s energy mostly from the tropics to extratropics. It is thus important to accurately measure and understand precipitation and its variability in order to properly understand the earth’s water and energy system.

Although there are currently more than 200 000 rain gauges actively measuring precipitation on land worldwide (Hulme 1992; New et al. 2001), their temporal and spatial distributions make it difficult to produce a consistent global land precipitation dataset for use in land hydrology. The advent of satellite precipitation retrieval technology has been the most effective solution to the problem and has been helping us to produce global precipitation datasets for more than two decades (e.g., Ferraro 1997; Ferraro et al. 2005; Huffman et al. 2009). Combining rain gauge measurement and satellite retrievals has been shown to give the most reliable solution for producing global precipitation datasets with good levels of spatial and temporal coverage (e.g., Xie and Arkin 1998; Huffman et al. 1997, 2009; Adler et al. 2003).

The primary source of global satellite precipitation retrievals over the past 20 years has been from microwave sensors on board various satellites. This includes the Special Sensor Microwave Imager (SSM/I) on the Defense Meteorological Satellite Program (DMSP) satellites (Ferraro 1997), the Advanced Microwave Sounding Unit (AMSU) on the National Oceanic and Atmospheric Agency (NOAA) Polar Orbiting Environmental Satellites (POES; Ferraro et al. 2005), the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI; Wang et al. 2009), and the Advanced Microwave Scanning Radiometer-Earth Observing System (AMSR-E) (Wilheit et al. 2003) on the National Aeronautics and Space Administration (NASA) and Japan Aerospace Exploration Agency (JAXA) joint satellites. The future international Global Precipitation Measurement (GPM) mission will provide better microwave precipitation retrievals and guarantee the continuation of global precipitation datasets (Hou 2006).

In general, microwave precipitation retrieval over land involves two steps: namely, the identification/screening of nonrainy pixels and the conversion of radiometer observations to rain rate (Ferraro et al. 1998). In the screening step, a decision is made for each pixel to determine whether the pixel views rainy conditions, or if other surfaces that have similar spectral signatures to rain are being observed. If rain conditions are determined, then channel information from satellite sensors is used to estimate the precipitation rate. Both steps have undergone periodic improvements over the past decade, based on feedbacks from the users of the global precipitation datasets derived from these microwave sensors.

The Grody–Ferraro screening methodology (Ferraro et al. 1986, 1998; Grody 1991) has been the most applied methodology for microwave land precipitation since its introduction. The methodology is built in within the Goddard profiling algorithm (GPROF; Kummerow et al. 2001), a widely used passive microwave precipitation retrieval algorithm. Various versions of GPROF have been applied in the SSM/I, TMI, and AMSR-E missions (McCollum and Ferraro 2003; Wang et al. 2009; Gopalan et al. 2010). An improved GPROF algorithm will also be applied for the GPM mission.

The current operational GPROF for TMI products version 6 (GPROFV6; McCollum and Ferraro 2003) and the newly improved GPROF version 7 (Gopalan et al. 2010) use a modified version-2 Goddard scattering algorithm (GSCAT-2; Adler et al. 1994) in their screening step. GSCAT-2 uses a fixed-grid land/sea mask and information from only 19-, 22-, and 85-GHz channels as inputs in defining rain and no-rain land regions. It was developed with channel information from SSM/I sensors on board the F8 DMSP satellite (Adler et al. 1994). This methodology requires several checks in its application, such as cold ocean and coastline checks (Adler et al. 1993), desert sand and cold surface checks (Grody 1991), and ambiguous cold surface-possible precipitation check (Adler et al. 1994). These checks are necessary because some surface effects may lead to the false identification of rain regions (Adler et al. 1994), such as over desert and snow- and ice-covered regions, and GSCAT-2 does not use any ancillary land surface cover information, which was not available at the time of its development. GSCAT-2 has been proven useful and performs well especially in the “SSM/I era” in defining rain and no-rain regions but not without some false rain identifications.

As the scientific community has increased its use of the level-2 (L2) and -3 (L3) land precipitation datasets, several “artifacts” that are related to misclassifications of rain and no-rain land regions have been noted, such as (positive and negative, depending on the seasons) biases in arid and semiarid land regions (i.e., Sahel, Kalahari), positive bias for inland water bodies (i.e., small lakes, large rivers), (again could be positive or negative depending on whether the algorithm detects more or less rain events, respectively) biases for snow-covered surfaces (i.e., melting snowpacks, high terrain snow, etc.), and seasonal biases over other regions (i.e., more or less detection of rain events). Details of these errors have been noted in many recent studies (e.g., Nesbitt et al. 2000; Zipser et al. 2006; Liu et al. 2008; Tian and Peters-Lidard 2007; Wang et al. 2009). In many instances, these problems look relatively small when inspecting the L2 swath retrievals; however, they tend to manifest themselves when averaged over longer time scales because they tend to persist in the same regions. They are also more noticeable when examining rain occurrences than rain amount because they sometimes result in light rain rates embedded within regions of heavier rain.

There are also issues with current screening methodologies that are related to how they were developed, such as the different field-of-view (FOV) sizes between TMI, AMSR-E, and SSM/I; the different spectral bands used by all of the sensors; and the vastly improved calibrations compared to those used in the early work for the F8 SSM/I. Thus, some of the thresholds in the screening are unreliable and are theoretically different from a physical standpoint due to varying FOV sizes and observation frequencies (and subsequently not portable from sensor to sensor). In particular, the differences arise at the interfaces of the various surface types, which sometimes rely on fixed thresholds, thus causing “ambiguous” classification and errors in the rainfall product.

Another limitation is the under-constrained problem of using strictly radiometric data; we cannot uniquely identify all types of land surfaces that are in the FOV using only satellite measurements. In today’s era of operational–routinely produced datasets, along with sufficient computing power, we can greatly improve the land surface classification for the GPM-era algorithms by utilizing highly accurate, reliable ancillary data—something that did not exist two decades ago. One final limitation of the current generation of land precipitation retrievals is that most of these algorithms rely on a fixed-grid land–sea–coast database that does not spatially match up with the satellite FOVs. These datasets also chose a grid spacing that is typically an “average” of the wide range of FOV sizes that exist on conical microwave sensors and does not necessarily reflect what the instrument is seeing as it enters into the algorithm and decision tree processing–classification. Additionally, the definition of “coast” may include relatively small inland water bodies or, conversely, exclude large inland water bodies, again offering “confusion” to the retrieval process. Finally, different land–sea–coast databases are currently used by the pre-GPM constellation members, presenting problems when trying to diagnose spurious precipitation regions in the blended products [e.g., TMI L3 (3B42) and NOAA Climate Prediction Center (CPC) morphing technique (CMORPH) data; e.g., Tian et al. (2007)].

To develop more physical approaches for land precipitation retrievals, all of these problem areas need to be addressed to advance the current state of the algorithms. This study presents a prototype, generic, unified screening methodology for microwave land precipitation retrieval algorithms that is applicable for all platforms and eliminates the need for the radiometric-only screening methodology that is used today.

2. Methodology development

A new generic, unified screening methodology for future microwave land precipitation retrieval algorithms should rely on ancillary datasets that are easily accessible, are stable, and accurately describe both the relatively static and dynamic features of the land surface cover. Such datasets have been made available partly through the advent of remotely sensed global land cover identification and characterization techniques in the last two decades. The use of these ancillary datasets will remove the current dependency of the screening step in the land precipitation retrieval algorithms and thus improve comparability between land precipitation products. This will eventually ensure the temporal continuity of global land precipitation datasets as satellite missions come and go with different periods of operation.

Figure 1 presents a schematic for the new unified screening methodology. The screening methodology starts with a satellite FOV that is used in a rain algorithm and in which surface characteristics are needed. With a spatial resolution dictated by the chosen FOV(s), the relatively static surface covers, such as land, water, land, water boundary (i.e., coast), desert, semiarid, and ice, are delineated from a static map of global surface cover types. Information on the climatological parameters of these surfaces may contribute to a better delineation. The final map of relatively static surface covers will then be combined with maps of dynamic surface covers, which may have different spatial and temporal resolutions depending on the characteristics of the available ancillary datasets. It is especially important to identify in the combined map those surface cover types that have been shown to have scattering signatures that are similar to rain in the radiometric measurements, such as desert, semiarid, snow, and ice (e.g., Ferraro et al. 1986; Grody 1991). The static and dynamic surface covers and climatological parameter information may come from remotely sensed global land cover datasets; remotely sensed and in situ measurements of physical properties, such as soil moisture, of the surface covers; and model-generated datasets such as the NASA Global Land Data Assimilation Systems (GLDAS; Rodell et al. 2004).

Fig. 1.
Fig. 1.

Schematic of the general structure of the new generic unified screening methodology for microwave land precipitation retrieval algorithms. Three land surface types are presented; the scheme can adopt more.

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

The FOV information will be used in preparing the static information of surface cover types. Here, we present a new method of aggregating information from ancillary surface cover maps. The new aggregation method is necessary because currently available surface covers maps have higher resolutions than the sizes of the FOV from any sensors for land precipitation retrievals. For example, the smallest FOV among currently operational satellites is 3.5 km × 5.9 km (89-GHz B channel of AMSR-E) and the spatial resolution of the University of Maryland Global Land Cover map is 1 km (UMD 1-km map; Hansen et al. 2000). Figure 2 (top) illustrates the current approach of aggregating land/water information in land precipitation retrieval algorithms. Essentially, each grid box in the land/water mask (red) contains aggregated information from the smaller grid boxes (blue) of the original land/water map. The information in the red grid box is only the percentage of land and remains as a fixed latitude/longitude grid, for example ⅙° in GPROF. This fixed grid cannot realistically represent the scan geometry and the varying FOV with frequency.

Fig. 2.
Fig. 2.

Aggregation methods of gridded information in the GPROFV6 with the (top) current and (bottom) prototype surface screening schemes.

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

In the new method (Fig. 2, bottom), all surface cover information (blue) from an original surface cover map and within a sensor resolution (red) is aggregated. The size of the red grid box represents an FOV. The aggregation information is stored in the center grid box (black). By doing so, we avoid averaging the higher-resolution original map into a fixed-grid lower-resolution map. The new aggregation method allows for the use of information from a center grid box (black) that is very close to an FOV and follows the swath of a scan. In the current method, an FOV may have to use information from the red grid box (Fig. 2, top), whose center is as far as half of the map’s grid diagonal (i.e., the red grid box’s diagonal). In the new aggregation method, the distance is less than the original map’s resolution.

A set of rules for use in assigning (i.e., flagging) a specific surface type to an FOV is required in the screening step. For example, GPROFV6 uses 0, 1, 2, and 3 for ocean, land, coast, and other surface types, respectively, for its land/water mask map. GPROFV6 will assign a value of 0 if the percentage of land is less than or equal to 2%. If the percentage of land is more than 95%, GPROFV6 will assign 1. If the percentage of land is more than 2% but less than or equal to 95%, GPROFV6 will assign 2. In the AMSU and SSM/I operational products, the rules for assigning surface types are done differently. Thus, it is important and preferable to have consistent, unified rules and surface-type numbers for all land precipitation retrieval algorithms in order to ensure compatibility and comparability between all land precipitation products. In today’s current era of L2 algorithms, there are inconsistencies on how this is done.

In this study, we propose to use the numbers 0, 1, 2, 3, and 4 for ocean, land, coast, desert, or semi/permanent snow/ice covered, and semiarid surface types, respectively. We will assign a value of 0 to an FOV if the percentage of water is more than 95%. If the percentage of water is more than 2% but less than or equal to 95%, we will assign 2. The dominant rule is applied for assigning 3 or 4; for example, if semiarid is the dominant surface type, then 4 will be assigned. However, complete information, on say, the top 5 surface types within an FOV of any given channel could also be retained if such information was deemed necessary by the algorithm developers. The rest of the surface will be assigned as 1.

Once the static land features are known, there are still many surface features that are very dynamic in nature and can vary on hourly to decadal time scales. Examples include snow and vegetation cover and soil moisture. All of these features impact the retrieval of precipitation over land. Another challenge is the effects of active precipitation on surface properties. Such information is crucial as input into the L2 retrieval algorithms.

The dynamic land features will be extracted from ancillary datasets and replace the static land features of the same FOV’s location. In this study, we screen out regions with altitudes at or above 2 km (obtained from the GPROFV6 surface elevation database; Fig. 3). In the future, however, we plan to replace the conservative rain detection removal in the high-elevation regions, mostly mountainous, with a detection scheme that is based on dynamic varying physical factors. The lack of ground truth validation data makes it hard to assess the reliability of precipitation retrievals in these regions. Also, because of the highly inhomogeneous spatial distribution and shadowing, the accuracy levels of the visible–infrared (VIS–IR) and microwave channels in detecting snow are diminished and not valid, respectively, in these regions.

Fig. 3.
Fig. 3.

Regions with elevation ≥2 km (blue).

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

3. Application and results

In this section, we present an application of the methodology developed in the previous section by using the GPROFV6 and TMI 1B11 datasets for producing L2 (2A12) data (GPROFV6 2A12). The UMD 1-km map is used as the static map along with the NOAA National Environmental Satellite, Data, and Information Service (NESDIS) global daily snow cover map from the combined NASA Moderate Resolution Imaging Spectroradiometer (MODIS) and AMSR-E data (Romanov et al. 2003).

There are a few global land cover maps available, such as the UMD 1-km map, the MODIS 1′ Ecosystem Classification Product (NASA 2009a), and the European Space Agency (ESA) Global Land Cover for 300 m (ESA 2009). The UMD 1-km map is attractive for use because it has been chosen for inclusion in GLDAS for its accuracy in depicting the global vegetation coverage (NASA 2009b). The future synergies between the satellite precipitation community and GLDAS are already being pursued. Additionally, the 1-km spatial resolution is fairly compatible with the FOV sizes of microwave radiometers; higher spatial resolution averaged to larger areas may not provide any additional information to algorithms. Also, it has 13 classes of vegetation cover (water, evergreen needle leaf forest, evergreen broadleaf forest, etc.), which might offer itself for stratification within the retrieval algorithm. The UMD 1-km map is based on the Advanced Very High Resolution Radiometer (AVHRR) 1-km data from 1 April 1992 through 31 March 1993. For our application, the file format of the UMD 1-km map is changed from the original GIS format to a binary format readable by the GPROFV6 2A12 algorithm.

Of the 13 classes of surface cover in the UMD 1-km map, we choose and apply water as water, bare ground as desert. and semi/permanent snow/ice-covered regions and open shrubland as semiarid regions in GPROFV6 2A12. Figure 4 shows various classes from the original UMD 1-km map for use in GPROFV6 2A12. By applying the new methodology, surface cover information from the UMD 1-km map is then aggregated for GPROFV6 2A12 with an FOV size of about 14 km × 14 km (Wang et al. 2009) (or equivalent to the 37-GHz channel on TMI). To give a place for the center grid box (the black box in Fig. 2, bottom) where all of the aggregation information will be stored, we use and aggregate information from a 15 km × 15 km box (the red box in Fig. 2, bottom).

Fig. 4.
Fig. 4.

The surface classes from the UMD 1-km Global Land Cover map to be used in GPROFV2 2A12 land precipitation retrievals with the prototype surface screening scheme: white, water; red, desert or semi/permanent snow/ice covered; yellow, semiarid; green, open grassland probably semiarid; and gray, other surface types.

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

Figure 5 shows the comparison between the current GPROFV6 2A12 surface-type map (top) and the surface-type map generated through the application of the new methodology (bottom) for TRMM orbit 42496 (1041:45–1214:08 UTC 30 April 2005). We can see that the new methodology better delineates the inland water and coastal regions and a thinner coast, which is advantageous because land and oceanic regions each uses a different precipitation detection and retrieval algorithm (Wilheit et al. 2003).

Fig. 5.
Fig. 5.

Comparison of using the (top) current, fixed-grid land/water mask and (bottom) prototype surface screening scheme via the UMD 1-km map and 1-km precision FOV mapping in GPROFV6 showing better inland water (blue) and coastal (yellow) regions (orbit 42496, at 1041:45–1214:08 UTC 30 Apr 2005).

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

In addition to the static UMD 1-km map, we use the NESDIS global snow cover map from the combined MODIS and AMSR-E dataset (Romanov et al. 2003) as the dynamic feature of land surface type. The map is generated by combining data from the VIS–IR and microwave channels. Snow detections with VIS–IR channels require daylight and will have gaps due to clouds. However, they have the highest spatial resolutions (0.5–5 km) and have a high probability of detection (around 95%). On the other hand, snow detection with microwave channels can be done under most weather conditions but with lower spatial resolution (25–50 km) and lower probability of detection (around 80%). Also, the microwave detection has problems with melting and shallow snows and frozen ground. The combined technique enables daily global all-weather snow detections at the highest spatial resolution (5° latitude/longitude resolution) starting from 9 June 2002.

Once the ancillary datasets are assembled as illustrated in Fig. 1, the GPROFV6 2A12 algorithm is run with TMI 1B11 data. Focusing on problematic cases of TMI 2A12 precipitation products over the United States, we are able to show improvements made through the application of the new methodology with the complete elimination of the GPROFV6 surface screening tests. Specifically, here we show the run for TRMM orbit 42496. We also compare the L2 precipitation products with precipitation retrievals from TRMM precipitation radar 2A25 (PR 2A25) for the orbits. Although PR 2A25 has narrower coincided swaths when compared to those of TMI 1B11, the PR 2A25 data are often used to validate TMI 2A12 precipitation products. Associated rain flag maps from the cases are also analyzed.

Figure 6 shows the GPROFV6 2A12 run for orbit 42496. Figure 6 (top) shows the precipitation retrieval of a very strong convective line over the United States by using the current GPROFV6 2A12 algorithm. Although the retrieval is comparable with PR 2A25 (Fig. 6, bottom), Fig. 6 (top) shows small regions (white) of no retrieval at the locations where the heaviest precipitation was occurring. The application of the new methodology removes these pixels (Fig. 6, middle) and hence results in better precipitation retrieval. Our examination of the rain flag map shows that the current algorithm detects cold (snow/ice) surfaces in these regions of no apparent retrieval (Fig. 7, top). We found that the very low values of brightness temperature at the 21-GHz vertical polarization (21V) channel triggered the algorithm to falsely identify the pixels as cold (snow/ice) surfaces. Hence, no retrieval was done over these pixels. The use of a high-resolution snow cover map, such as the NESDIS global snow cover map, makes it possible to correct the false cold surface detection in the new algorithm (Fig. 7, bottom). Figure 7 (bottom) also shows the false cold surface detection over the inland coast (see Fig. 6, bottom). Because of the application of higher-resolution static surface maps (i.e., the UMD 1-km map), many small inland waters are detected in the new static surface screen, as are their coasts. GPROFV6 2A12 has a separate algorithm in the screening step for coastal regions. Because we do not develop and maintain the ocean and coast algorithms of GPROFV6 2A12, we did not change the algorithms for the application of the new methodology.

Fig. 6.
Fig. 6.

The GPROFV6 2A12 retrievals with the (top) current and (middle) prototype surface screening schemes and (bottom) PR 2A25 retrievals. Regions of no retrieval are in white (orbit 42496, 1041:45–1214:08 UTC 30 Apr 2005).

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

Fig. 7.
Fig. 7.

The GPROFV6 2A12 rain flag maps with the (top) current and (bottom) prototype surface screening schemes. False snow detections (green) are removed in the prototype scheme. Also shown are rain (red) and no-rain land (pink) and water–oceanic regions (blue) (orbit 42496, 1041:45–1214:08 UTC 30 Apr 2005).

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

The application of the new surface identification scheme changes the number of detected rain events in the GPROFV6 2A12. Figure 8 shows the differences in the total numbers of detected rain events with rain rates of more than 0 mm h−1 between the current and new GPROFV6 2A12 runs within a 38°S–38°N belt in the summer and winter seasons. In summer, defined here as the months of June–August (JJA) 2008 for the Northern Hemisphere and December 2008–February 2009 (DJF) for the Southern Hemisphere, the application of the static global land cover UMD map in GPROFV6 2A12 conservatively removes rain detections over desert, semiarid, and snow- and ice-covered regions. For example, noticeably fewer nonzero rain events are detected in the new GPROFV6 2A12 over desert and semiarid regions (such as the African Sahara, Sahel, Namib, and Kalahari and Australian Great Sandy regions; see the blue colors in Fig. 8). Conversely, in strong convective areas (such as the Amazon), the new scheme eliminates false removals of rain events, resulting in more rain detections in these heavy rainfall zones (red colors in Fig. 8).

Fig. 8.
Fig. 8.

Nonzero rain event detection comparisons between the current and proposed surface screening scheme applications in GPROFV6 2A12 in JJA and DJF. Values are (proposed GPROFV6 2A12 − current GPROFV6 2A12) from total rain events with rain >0 mm h−1.

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

In winter, defined here as the months of JJA 2008 for the Southern Hemisphere and DJF 2009 for the Northern Hemisphere, the new GPROFV6 2A12 noticeably detects more nonzero rain events over cold regions such as the eastern United States, the great plain of China, and the southern parts of the Australian and African continents and South America (red colors in Fig. 8). As shown earlier, the application of the NESDIS global snow cover map in the proposed GPROFV6 2A12 removes false snow- or ice-cover detections in the current GPROFV6 2A12. More nonzero rain events are also detected over regions in the vicinity of the Sahel. Rain detections are allowed over the regions because they are not classified as being arid or semiarid in the UMD 1-km map. This demonstrates that a more seasonally driven surface-type classification is needed in these regions. Such datasets are readily available for use but are beyond the scope of this study.

In both the summer and winter seasons, because the new GPROFV6 2A12 does not detect rain over the regions with altitudes at or above 2 km (Fig. 3), fewer nonzero rain events are detected over the high-elevation regions (such as the Iranian and Himalayan–Tibetan Plateaus, the Rockies, and the Andes regions) (blue colors; Fig. 8).

Another type of analysis that was performed to study the impacts of the new surface identification scheme is through a comparison with the TRMM 2A25 rainfall. Figure 9 shows normalized biases between the current and new GPROFV6 2A12 and PR 2A25 runs over the regions where there are differences in the retrievals between the current and new GPROFV6 2A12 within the 38°S–38°N belt in the summer and winter seasons. Figure 9 shows the difference between the current and new GPROFV6 2A12 and PR 2A25 runs (GPROFV6 2A12 − PR 2A25) normalized by PR 2A25. Rain rates (mm day−1) averaged over each season are used in the computations. Summer precipitation comparisons between GPROFV6 2A12 and PR 2A25 show that the application of the new methodology, particularly the application of the static global land cover UMD map in GPROFV6 2A12, results in underestimations over desert, semiarid, and snow- and ice-covered regions, such as the African Sahara, Sahel, Namib, and Kalahari regions and the Australian Great Sandy region (blue colors; in Fig. 9). Underestimations by the new GPROFV6 2A12 (blue colors in Fig. 9) are also noticeable over the high-elevation regions (with altitudes at or above 2 km), such as the Iranian and Himalayan–Tibetan Plateaus, the Rockies, and the Andes regions. These patterns of underestimations over the desert, semiarid, snow- and ice-covered, and high-elevation regions by the new GPROFV6 2A12 in Fig. 9 (blue colors) are similar to the patterns of fewer nonzero rain event detections by the new GPROFV6 2A12 over the regions in Fig. 8 (blue colors). The similarity shows that most summer precipitation overestimations (red colors in Fig. 9) by the current GPROFV6 2A12 over these regions can be associated with the screening step in the algorithm.

Fig. 9.
Fig. 9.

Comparisons between GPROFV6 2A12 and PR 2A25 in the regions where there are differences in the retrievals between the GPROFV6 2A12 with the current and proposed surface screening schemes in JJA and DJF. Values are (GPROFV6 2A12 − PR 2A25)/PR 2A25 from rain rates (mm day−1) averaged in each season.

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

The spatial correlations between the current and new GPROFV6 2A12 and PR 2A25 over the eastern U.S. land region (15°–38°N, 95°–75°W) in summer (JJA) are 0.61 and 0.58, respectively. Because the eastern U.S. land area does not have desert, semiarid, snow- and ice-covered, or high-elevation (with altitudes at or above 2 km) regions, changes in land/water delineation by the new GPROFV6 2A12 may contribute to the slight differences in the correlations. We do not include the rest of U.S. landmass in the correlation analysis because of the removal of desert, semiarid, snow- and ice-covered, and high-elevation regions. The removal will result in unfair comparisons between the current and new GPROFV6 2A12 results.

In winter, overestimations by the new GPROFV6 2A12 are noticeable over cold regions, such as the northern part of the United States and the great plain of China, as well as the southern parts of Australia, Africa, and South America (red colors in Fig. 9). False snow or ice cover detections is one of the problems in the current GPROFV6 2A12, as shown earlier in this section. These false detections prevent the current GPROFV6 2A12 from moving forward to the conversion step and hence no retrieval is done. Again, comparisons of these patterns with patterns of more detections by the new GPROFV6 2A12 in Fig. 8 show that most winter precipitation underestimations by the current GPROFV6 2A12 over these regions can be associated with the screening step in the algorithm.

Patterns of underestimations over the high-elevation regions (with altitudes at or above 2 km) by the new GPROFV6 2A12 in winter in Fig. 9 (blue colors) are similar with the patterns of fewer nonzero rain event detections by the new GPROFV6 2A12 over the regions in Fig. 8 (blue colors). Again, the similarity shows that most summer precipitation overestimations (red colors in Fig. 9) by the current GPROFV6 2A12 over these regions can be associated with the screening step in the algorithm.

In winter (DJF), we detect a slight improvement in the spatial correlations between the current (0.53) and new (0.55) GPROFV6 2A12 and PR 2A25 results over the eastern U.S. land region (15°–38°N, 95°–75°W). The removal of false snow detections in the new GPROFV6 2A12 may contribute to the improvement in the correlations.

The new GPROFV6 2A12 also shows improvements in detecting inland waters and coastal regions and hence adjacent land and oceanic regions. Figure 8 shows the Amazon River and its tributaries implicitly through changes in nonzero rain event detections by the new algorithm. Changes are also shown in regions with many islands such as the Indonesian Archipelago and the Caribbean, Madagascar, the Philippines, Japan, and the Solomon and Polynesian, Canary, and Hawaiian Islands (Fig. 9).

Seasonal comparisons of the recently released Global Precipitation Climatology Project (GPCP), version 2.1, (GPCPV2.1; Huffman et al. 1997, 2009; Adler et al. 2003) with the current and new GPROFV6 2A12 are also done over the U.S. region only. The region is chosen because the GPCPV2.1 uses gauge data from the improved Global Precipitation Climatology Centre (GPCC) precipitation gauge analyses (Huffman et al. 2009) for land precipitation analysis and the United States has the best rain gauge coverage when compared to the other regions in the 38°S–38°N belt. Hence, the GPCP land precipitation over the United States is more reliable than over the other regions in the latitudinal belt. Figure 10 shows normalized biases between the current and new GPROFV6 2A12 and GPCPV2.1 runs in the summer (JJA) and winter (DJF) seasons. Figure 10 shows the difference between the current and new GPROFV6 2A12 and GPCPV2.1 (GPROFV6 2A12 − GPCPV2.1) normalized by GPCPV2.1. Again, rain rates (mm day−1) averaged over each season are used in the computations. In summer (JJA), smaller biases (lighter red and especially blue colors) are noticeable over the western United States (125°–95°W) after the application of the new methodology (Fig. 10, bottom left). These changes are especially significant over the regions where the 2-km high-elevation cutoff is applied in the new GPROFV6 2A12 (see Fig. 3). The biases over the eastern United States (95°–75°W) are not noticeably changed from the current to the new GPROFV6 2A12 runs.

Fig. 10.
Fig. 10.

Comparisons between GPCPV2.1 and GPROFV6 2A12 with the (top) current and (bottom) prototype surface screening schemes in (left) JJA and (right) DJF. Values are (GPROFV6 2A12 − GPCPV2.1)/GPCPV2.1 from rain rates (mm day−1) averaged in each season.

Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2572.1

The spatial correlations between the current and new GPROFV6 2A12 and GPCPV2.1 over the eastern U.S. land region (95°–75°W) in summer (JJA) are 0.81 and 0.78, respectively. These correlations are much different from the correlations between the current (0.61) and new (0.58) GPROFV6 2A12 and PR 2A25 simulations. We suspect that the use of much coarser resolution (2.5) datasets in computing the later correlation may have contributed to its higher value by masking finer differences between the datasets. The correlations with PR 2A25 was computed by using 0.25°-resolution datasets.

In winter (DJF), the new GPROFV6 2A12 noticeably detects more rain over the eastern United States, overcoming the false snow or ice cover detection problem in the current GPROFV6 2A12, as previously shown in this section. This results in better biases from the new GPROFV6 2A12 (more lighter blue or red colors in Fig. 10, bottom right). This improvement is especially important because a conservative rain detection removal is not done over the region. This means that the use of an ancillary snow dataset improves rain detection in the region during winter. The spatial correlation between the new GPROFV6 2A12 and GPCPV2.1 (0.71) over the eastern U.S. land region (95°–75°W) is the same as the correlation between the current GPROFV6 2A12 and GPCPV2.1 (0.71).

4. Conclusions

We have developed a new generic, unified screening methodology for GPM-era microwave land precipitation retrieval algorithms by using ancillary datasets as an alternative to the current “surface screening” approach, which relies on a fixed land/sea mask and radiometric measurements. Our early prototype of the new surface screening scheme has shown that the methodology is promising and removes the current dependency of the screening step on relatively different satellite specific channels. The new ancillary data approach is technologically feasible because of the advent of remotely sensed global land cover identification and characterization techniques and the advancement of computing and data storage technology, which did not exist two decades ago. The approach will eventually ensure the comparability and continuity of satellite-based precipitation products. The new method is flexible in that it can incorporate any set of ancillary data, both static and dynamic in nature.

Our experiment has shown that many noticeable anomalies in the GPROFV6 2A12 land precipitation retrievals, such as overestimation in summer over desert regions and underestimation in winter over cold surfaces, are related to misclassifications of land surface types in the screening step. This misclassification leads to false rain detections and ultimately erroneous precipitation retrievals. By dynamically using ancillary static and dynamic information of the land surface types, we have been able to show improvements in rain detection and hence better retrievals of precipitation in the problematic desert, semiarid, snow- and ice-covered, and coastal and inland water regions, where incorrect removals of valid raining events and misidentifications of surfaces that are similar to rain are not done. Outside these regions, precipitation retrievals and amounts are unaffected by the new surface screening methodology.

We plan to further develop our prototype of the new surface screening scheme by adding more static and dynamic information from remotely sensed global land cover datasets, remotely sensed and in situ measurements of physical properties, and model-generated datasets such as the reanalyses and the NASA Global Land Data Assimilation Systems (GLDAS; Rodell et al. 2004). Dynamic varying factors, such as desertification, deforestation, greening, atmospheric conditions (such as freezing-level heights and thickness and warm layer), soil wetness, flooding, etc., that will effect the scattering signatures of any surface covers should all be considered and taken into account in developing the prototype. By doing so, conservative rain detection removal, such as the one that is currently being applied in our prototype for high-elevation (more than 2 km high) regions, can be replaced by a more dynamic scheme. This will ultimately result in better rain detection and retrieval. Sensitivity analyses of the temporal and spatial resolutions of the ancillary data and various FOV sizes on retrieval errors should also be analyzed. Further development will also be geared toward its applications on different current and future satellite precipitation retrieval platforms and integrations with models such as GLDAS.

Acknowledgments

The authors gratefully thank Christa D. Peters-Lidard and Daniel J. Cecil for their input to this study. This research is supported under NOAA Grant NA17EC14 83 to ESSIC/UMCP. The participation of AS is supported by NOAA/NESDIS and ITB. The participation of NYW and KG is supported by NOAA/NESDIS. The views, opinions, and findings contained in this report are those of the authors and should not be construed as an official NOAA, ITB, or U.S. or Indonesian government position, policy, or decision.

REFERENCES

  • Adler, R. F., , A. J. Negri, , P. R. Keehn, , and I. M. Hakkarinen, 1993: Estimation of monthly rainfall over Japan and surrounding waters from a combination of low-orbit microwave and geosynchronous IR data. J. Appl. Meteor., 32, 335356.

    • Search Google Scholar
    • Export Citation
  • Adler, R. F., , G. J. Huffman, , and P. R. Keehn, 1994: Global tropical rain rate estimates from microwave-adjusted geosynchronous IR data. Remote Sens. Rev., 11, 125152.

    • Search Google Scholar
    • Export Citation
  • Adler, R. F., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 11471167.

    • Search Google Scholar
    • Export Citation
  • ESA, cited 2009: ESA/ESA GlobCover Project, led by MEDIAS-France. [Available online at http://ionia1.esrin.esa.int/index.asp.]

  • Ferraro, R., 1997: SSM/I derived global rainfall estimates for climatological applications. J. Geophys. Res., 102, 16 71516 735.

  • Ferraro, R., , N. C. Grody, , and J. A. Kogut, 1986: Classification of geophysical parameters using passive microwave satellite measurements. IEEE Trans. Geosci. Remote Sens., GE-24, 10081013.

    • Search Google Scholar
    • Export Citation
  • Ferraro, R., , E. A. Smith, , W. Berg, , and G. Huffman, 1998: A review of screening techniques for passive microwave precipitation retrieval algorithms. J. Atmos. Sci., 55, 15831600.

    • Search Google Scholar
    • Export Citation
  • Ferraro, R., and Coauthors, 2005: NOAA operational hydrological products derived from the AMSU. IEEE Trans. Geosci. Remote Sens., 43, 10361049.

    • Search Google Scholar
    • Export Citation
  • Gopalan, K., , N.-Y. Wang, , R. Ferraro, , and C. Liu, 2010: Status of the TRMM 2A12 land precipitation algorithm. J. Atmos. Oceanic Technol., 27, 13431354.

    • Search Google Scholar
    • Export Citation
  • Grody, N. C., 1991: Classification of snow cover and precipitation using the Special Sensor Microwave/Imager (SSM/I). J. Geophys. Res., 96, 74237435.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., and Coauthors, 2000: Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21, 13311364.

    • Search Google Scholar
    • Export Citation
  • Hou, A. Y., 2006: The Global Precipitation Measurement (GPM) mission: An overview. 2006 EUMETSAT Meteorological Satellite Conf, Helsinki, Finland, EUMETSAT, P.48. [Available online at http://gpm.gsfc.nasa.gov/science/sciencetalks/Hou_EUMETSAT06.pdf.]

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 520.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 2009: Improving the global precipitation record: GPCP version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009GL040000.

    • Search Google Scholar
    • Export Citation
  • Hulme, M., 1992: Rainfall changes in Africa: 1931–60 to 1961–90. Int. J. Climatol., 12, 685699.

  • Kummerow, C., and Coauthors, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 18011820.

    • Search Google Scholar
    • Export Citation
  • Liu, C., , E. Zipser, , D. Cesil, , S. Nesbitt, , and S. Sherwood, 2008: A cloud and precipitation feature database from 9 years of TRMM observations. J. Appl. Meteor. Climatol., 47, 27122728.

    • Search Google Scholar
    • Export Citation
  • McCollum, J., , and R. R. Ferraro, 2003: The next generation of NOAA/NESDIS SSM/I, TMI and AMSR-E microwave land rainfall algorithms. J. Geophys. Res., 108, 83828404.

    • Search Google Scholar
    • Export Citation
  • NASA, cited 2009a: NASA/MODIS Atmosphere Program. [Available online at http://modis-atmos.gsfc.nasa.gov/ECOSYSTEM/index.html.]

  • NASA, 2009b: NASA/Land Data Assimilation Systems. [Available online at http://ldas.gsfc.nasa.gov.]

  • Nesbitt, S. W., , E. J. Zipser, , and D. J. Cecil, 2000: A census of precipitation features in the tropics using TRMM: Radar, ice scattering, and lightning observations. J. Climate, 13, 40874106.

    • Search Google Scholar
    • Export Citation
  • New, M., , M. Todd, , M. Hulme, , and P. Jones, 2001: Review precipitation measurements and trends in the twentieth century. Int. J. Climatol., 21, 18991922.

    • Search Google Scholar
    • Export Citation
  • Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381394.

  • Romanov, P., , D. Tarpley, , G. Gutman, , and T. Carroll, 2003: Mapping and monitoring of the snow cover fraction over North America. J. Geophys. Res., 108, 8619, doi:10.1029/2002JD003142.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., , and C. D. Peters-Lidard, 2007: Systematic anomalies over inland water bodies in satellite-based precipitation estimates. Geophys. Res. Lett., 34, L14403, doi:10.1029/2007GL030787.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., , C. D. Peters-Lidard, , B. J. Choudhury, , and M. Garcia, 2007: Multitemporal analysis of TRMM-based satellite precipitation products for land data assimilation applications. J. Hydrometeor., 8, 11651183.

    • Search Google Scholar
    • Export Citation
  • Wang, N.-Y., , C. Liu, , R. Ferraro, , D. Wolff, , E. Zipser, , and C. Kummerow, 2009: TRMM 2A12 land precipitation product – Status and future plans. J. Meteor. Soc. Japan, 87, 237253.

    • Search Google Scholar
    • Export Citation
  • Wilheit, T. A., , C. Kummerow, , and R. Ferraro, 2003: Rainfall algorithms for AMSR-E. IEEE Trans. Geosci. Remote Sens., 41, 204214.

  • Xie, P., , and P. A. Arkin, 1998: Global monthly precipitation estimates from satellite-observed outgoing longwave radiation. J. Climate, 11, 137164.

    • Search Google Scholar
    • Export Citation
  • Zipser, E., , D. Cecil, , C. Liu, , S. Nesbitt, , and S. Yuter, 2006: Where are the most intense thunderstorms on Earth? Bull. Amer. Meteor. Soc., 87, 10571071.

    • Search Google Scholar
    • Export Citation
Save