The accuracy of three widely-used,
near-global, high-resolution satellite rainfall products (CMORPH, TMPA-RT v7,
TMPA-RP v7) is assessed using relatively dense networks of rain gauges deployed
at two experimental sites that represent extreme topographic features of the
Blue Nile River Basin: (1) low-elevation grid located in the lowland plain of
the basin, and (2) high-elevation grid located in the highland mountainous
region of the basin. Results show that the accuracy of satellite rainfall estimates depends on rainfall rate, underlying
topography, and retrieval algorithm. The
satellite estimates produce a positive bias for light rainfall and negative
bias for moderate and heavy rainfall, and are less capable of detecting rainfall at the high-elevation site.
CMORPH and TMPA-RT overestimate
mean rainfall at the low-elevation site but underestimate it at the
high-elevation site. TMPA-RT and TMPA-RP underestimate frequency but
overestimate intensity of rainfall in both regions. Of all the products, CMORPH shows superior
performance in estimating the temporal fluctuation of rainfall, detecting rain,
and capturing the diurnal cycle of rainfall. However, CMORPH is the most biased
estimate. TMPA-RT and TMPA-RP provide lower bias, but this comes as the result
of two substantially large errors that tend to cancel each other while
computing the bias: substantial underestimation of rainfall occurrence and
substantial overestimation of rain intensity. Although the TMPA-RP estimates are
primarily developed to remove the bias and improve the overall accuracy of the
TMPA-RT estimates, the TMPA-RP estimates show by far inferior performance than
the TMPA-RT estimates by all accounts including bias. Of all three satellite
rainfall products considered, TMPA-RP shows the worst and unacceptable
performance by all accounts.