IEEE Geoscience and Remote Sensing Magazine - June 2014 - 31

15 Oct 2006

10 Jul 2009

01 Feb 2010

CMORPH

PERSIANNCCS

Merged

3B42RT

05 Jan 2005

mm/day
0.1

100

125

150

Figure 5. Comparison of Performance Based Merged Product (PBMP) and individual satellite rainfall products at 0.25 degree spatial and a
day temporal resolution for randomly selected days.

is based on runoff error variance which is estimated from
the regression model as discussed in section 3. The extent of
the merged product is, therefore, limited to the land portion
of the earth's surface. The merged product is now available
at spatial resolution of 0.25 degree between 50c north and
south latitude and temporal scale of a day for the period of
2003-2010. For the interested readers and due to size limitations, we have made publicly available only a subset of this
merged dataset spanning May 2003 to July 2003 at http://
iweb.tntech.edu/fhossain/papers/MergedPrecipDataSample.zip. However, the entire 8 years (2003-2010) merged
data (8 GB in size) can be provided on request to any of the
authors via ftp or media mail.
As seen in Fig. 5, the magnitude and spatial rainfall distribution pattern of the three satellite rainfall products are
not consistently the same. Undeniably, the 3B42RT and
CMORPH show a close similarity in all selected days. The
PERSIANN-CCS has a different feature both in magnitude
and spatial pattern than the other two products. Therefore, the question is how well does the merged product
derived from the three satellite rainfall estimates capture
the magnitude and spatial pattern of the ground truth
data in selected test sites?
Fig. 6(a) shows the POD of the original and merged
satellite rainfall products over the study regions. The POD
signifies the percentage of rainfall events that are detected
correctly. In all study regions and all seasons, the merged
product exhibits high value of POD (almost greater than
0.6, except for summer season in Middle East region). This
shows that in terms of the spatial accuracy, the merged product is the best estimates from the original products. High
june 2014

ieee Geoscience and remote sensing magazine

POD implies that the spatial distribution of the rainfall pattern is well spotted by the algorithm during the screening
stage of the retrieval process. But it does not fully indicate
the level of accuracy in estimation of the rainfall magnitude.
Fig. 6(b) demonstrates a false alarm categorical verification measure for rainfall forecast that is associated
with no rain observed. The FAR represents the fraction
of detected rainy grid cells that were found to be nonrainy. The merged product has the highest FAR next to
PERSIANN-CCS product.
As it is a combined signature
of three satellite rainfall
The merging concepT is
products, high FAR should
designed To exploiT The
be expected in the merged
complemenTary
product. As seen from the
result, most of the false sigadvanTage oF each
nal within PERSIANN-CCS
rainFall esTimaTion
estimate is also inherent
algoriThm.
into the merged product.
Like POD, FAR is also associated with the spatial accuracy but not with the accuracy of the estimated rainfall
magnitude.
The next verification measure of spatial accuracy is
the Threat Score (TR), which represents the fraction of
detected rain event that were detected correctly. TR is also
called Critical Success Index (CSI) with an ideal value of
one. The CSI is not affected by the number of non-event
forecasts that verify a no-event forecast associated with no
event observed. In case of large number of miss and false
precipitation events, the CSI value is generally low. Fig 6(c)
31

http://http:// http://iweb.tntech.edu/fhossain/papers/MergedPrecipDataSam

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2014