388
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
instruments into space (Hou et al. 2008), makes satellite precipitation estimates
suitable for hydrologic applications. It is expected that, through the deployment of
the Global Precipitation Measurement (GPM) Mission (Hou et al. 2008), satellite
precipitation observations at every location around the world every 2–4 h will be
globally collected (Hou et al. 2008). The high temporal sampling may significantly
reduce sampling errors to a level that is acceptable for hydrologic applications. In
addition to small sampling error, improvements in satellite precipitation estimates
are expected in the GPM era. These improvements will be achieved through the
use of high-quality physically consistent databases of precipitation and associated
brightness temperatures. That is, given the multitude of precipitation profiles that
can be associated with an instantaneous set of brightness temperatures at a given
location, a priori information is required to derive unique estimates from satellite
radiometer observations. Bayesian formulations (Kumerrow et al. 1996; Pierdicca et
al. 1996) have been extremely popular in the last 15 years in deriving unique physically acceptable precipitation retrievals from satellite radiometer observations. In
Bayesian formulations, given a set of instantaneous brightness temperature observations at a specific location, search procedures are used to determine the precipitation
profiles in the database that are characterized by brightness temperatures similar
to the actual observations, and a composite solution is determined on the basis of
these profiles. Although other retrieval formulations are possible (e.g., the reduction
of the number of variables through statistical procedures to a small set that can be
determined through inverse radiative transfer modeling), Bayesian formulations are
preferable due to their computational efficiency and relative simplicity.
Before the deployment of the Tropical Rainfall Measuring Mission (TRMM)
(Simpson et al. 1996), the construction of databases to support Bayesian precipitation
retrievals from satellite radiometer observations relied on cloud-resolving models
(CRM) (Kumerrow et al. 2001). However, systematic differences between TRMM
Precipitation Radar (PR) estimates and TRMM Microwave Imager (TMI) estimates
derived using databases constructed from CRM simulations were noted (Adler
et al. 2000) and were interpreted in various studies (Masunaga and Kummerow
2005; Grecu and Olson 2006) as potential limitations in CRM-based databases. The
existence of coincident TRMM PR and TMI observations make possible the development of precipitation brightness temperatures databases directly from observations (Masunaga and Kummerow 2005; Grecu and Olson 2006). Such databases
were proven to significantly reduce the systematic difference TRMM PR and TMI
estimates over oceans (Grecu and Olson 2006). Although overland radiometer precipitation retrieval algorithms have been derived from coincident TRMM PR and
TMI observations (Grecu and Anagnostou 2001), such algorithms are applicable
only to TMI-like instruments and therefore not appropriate for the derivation of
unified satellite precipitation estimates from a variety of instruments envisioned by
Hou et al. (2008). To derive unified satellite precipitation over land, the TRMM PR
and TMI observations have to be converted into precipitation and other variables
to which the satellite radiometer observations are sensitive (e.g., nonprecipitating
cloud, water vapor, surface emissivity, or surface temperature), from which satellite brightness temperatures can be calculated at frequencies and resolutions specific to other existent satellite radiometers. Such conversions, commonly known as
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
instruments into space (Hou et al. 2008), makes satellite precipitation estimates
suitable for hydrologic applications. It is expected that, through the deployment of
the Global Precipitation Measurement (GPM) Mission (Hou et al. 2008), satellite
precipitation observations at every location around the world every 2–4 h will be
globally collected (Hou et al. 2008). The high temporal sampling may significantly
reduce sampling errors to a level that is acceptable for hydrologic applications. In
addition to small sampling error, improvements in satellite precipitation estimates
are expected in the GPM era. These improvements will be achieved through the
use of high-quality physically consistent databases of precipitation and associated
brightness temperatures. That is, given the multitude of precipitation profiles that
can be associated with an instantaneous set of brightness temperatures at a given
location, a priori information is required to derive unique estimates from satellite
radiometer observations. Bayesian formulations (Kumerrow et al. 1996; Pierdicca et
al. 1996) have been extremely popular in the last 15 years in deriving unique physically acceptable precipitation retrievals from satellite radiometer observations. In
Bayesian formulations, given a set of instantaneous brightness temperature observations at a specific location, search procedures are used to determine the precipitation
profiles in the database that are characterized by brightness temperatures similar
to the actual observations, and a composite solution is determined on the basis of
these profiles. Although other retrieval formulations are possible (e.g., the reduction
of the number of variables through statistical procedures to a small set that can be
determined through inverse radiative transfer modeling), Bayesian formulations are
preferable due to their computational efficiency and relative simplicity.
Before the deployment of the Tropical Rainfall Measuring Mission (TRMM)
(Simpson et al. 1996), the construction of databases to support Bayesian precipitation
retrievals from satellite radiometer observations relied on cloud-resolving models
(CRM) (Kumerrow et al. 2001). However, systematic differences between TRMM
Precipitation Radar (PR) estimates and TRMM Microwave Imager (TMI) estimates
derived using databases constructed from CRM simulations were noted (Adler
et al. 2000) and were interpreted in various studies (Masunaga and Kummerow
2005; Grecu and Olson 2006) as potential limitations in CRM-based databases. The
existence of coincident TRMM PR and TMI observations make possible the development of precipitation brightness temperatures databases directly from observations (Masunaga and Kummerow 2005; Grecu and Olson 2006). Such databases
were proven to significantly reduce the systematic difference TRMM PR and TMI
estimates over oceans (Grecu and Olson 2006). Although overland radiometer precipitation retrieval algorithms have been derived from coincident TRMM PR and
TMI observations (Grecu and Anagnostou 2001), such algorithms are applicable
only to TMI-like instruments and therefore not appropriate for the derivation of
unified satellite precipitation estimates from a variety of instruments envisioned by
Hou et al. (2008). To derive unified satellite precipitation over land, the TRMM PR
and TMI observations have to be converted into precipitation and other variables
to which the satellite radiometer observations are sensitive (e.g., nonprecipitating
cloud, water vapor, surface emissivity, or surface temperature), from which satellite brightness temperatures can be calculated at frequencies and resolutions specific to other existent satellite radiometers. Such conversions, commonly known as