We explore the potential of spaceborne radar (SR) observations
from the Ku-band precipitation radars onboard the Tropical Rainfall Measuring
Mission (TRMM) and Global Precipitation Measurement (GPM)
satellites as a reference to quantify the ground radar (GR) reflectivity
bias. To this end, the 3-D volume-matching algorithm proposed by
Weather radars are essential tools in providing high-quality information
about precipitation with high spatial and temporal resolution in three
dimensions. However, several uncertainties deteriorate the accuracy of
rainfall products, with calibration contributing the most amount
Since both ground radars and spaceborne precipitation radars provide a
volume-integrated measurement of reflectivity, a direct comparison of the
observations can be done in three dimensions
Relative calibration between SRs and GRs was originally suggested by
Due to different viewing geometries, ground radars and spaceborne radars are affected by different sources of uncertainty and error. Observational errors with regard to atmospheric properties such as reflectivity are, for example, caused by ground clutter or partial beam blocking. Persistent systematic errors in the observation of reflectivity by ground radars are particularly problematic: the intrinsic assumption of the bias estimation is that the only systematic source of error is radar calibration. It is therefore particularly important to address such systematic observation errors.
In this study, we demonstrate that requirement with the example of partial
beam blocking. The analysis is entirely based on algorithms implemented in
the wradlib open-source software library
Precipitation radar data were gathered from TRMM 2A23 and 2A25 version 7
products
It is important to note that, at the time of writing, changes in calibration
parameters applied in the GPM Version 5 products resulted in an increase of
+1.1 dB from the corresponding TRMM version 7 products
The Philippine Atmospheric, Geophysical, and Astronomical Services Administration (PAGASA) maintains a nationwide network of 10 weather radars, 8 of which are single-polarization S-band radars and 2 dual-polarization C-band radars. The Subic radar, which covers the greater Metropolitan Manila area, has the most extensive set of archived data. The radar coverage includes areas that receive some of the highest mean annual rainfall in the country.
The Subic radar sits on top of a hill at 532 m a.s.l. in the municipality
of Bataan, near the border with Zambales (location: 14.82
Characteristics of the Subic radar and its volume scan strategy. The numbers in parentheses correspond to scans in 2015, where the scanning strategy was different due to hardware issues.
In an ideal situation, SR and GR should have the same measurements for the same volume of the atmosphere, as they are measuring the same target. However, observational differences may arise due to different view geometries, different operating frequencies, different environmental conditions of each instrument, and different processes along the propagation path of the beam. As pointed out before, we focus on beam blockage as an index of GR data quality.
In regions of complex topography, ground radars are typically affected by the
effects of beam blockage, induced by the interaction of the beam with the
terrain surface resulting in a weakening or even loss of the signal. To
quantify that process within the Subic radar coverage, a beam blockage map is
generated following the algorithm proposed by
The beam blockage fraction was calculated for each bin and each antenna pointing angle. The cumulative beam blockage was then calculated along each ray. A cumulative beam blockage fraction (BBF) of 1.0 corresponds to full occlusion, and a value of 0.0 to perfect visibility.
The quality index based on beam blockage fraction is then computed following
A slightly different formulation to transform partial beam blockage to a
quality index has been presented in other studies
Figure
Quality index map of the beam blockage fraction for the Subic radar
at
As expected, the degree of beam blockage decreases with increasing antenna
elevation, yielding the most pronounced beam blockage at 0.0
SR and GR data were matched only for the wet period within each year, which
is from June to December. Several meta-data parameters were extracted from
the TRMM 2A23/2A25 and GPM 2AKu products for each SR gate, such as the
corresponding ray's bright band height and width, gate coordinates in three
dimensions (longitude and latitude of each ray's Earth intercept and range
gate index), time of overpass, precipitation type (stratiform,
convective, or other), and rain indicators (rain certain or no-rain).
The parallax-corrected altitude (above mean sea level) and horizontal
location (with respect to the GR) of each gate were determined as outlined in
the appendix of
For each SR overpass, the GR sweep with the scan time closest to the overpass
time within a 10 min window (
Diagram illustrating the geometric intersection.
In order to minimize systematic differences in comparing the SR and GR
reflectivities caused by the different measuring frequencies, the SR
reflectivities were converted from Ku- to S-band following the formula
The actual volume matching algorithm closely follows the work of
The nominal minimum sensitivity of both TRMM PR and GPM KuPR is 18 dBZ, so
only values above this level were considered in the calculation of average SR
reflectivity in the matched volume. In addition, the fraction of SR gates
within a matched volume above that threshold was also recorded. On the other
hand, all GR bins are included in the calculation of average GR reflectivity,
after setting the bins with reflectivities below 0.0 to 0.0 dBZ, as
suggested by
Filtering criteria for the matching workflow.
Beam blockage and the corresponding GR quality maps were computed for each GR
bin (cf. Sect.
To analyze the effect of data quality on the estimation of GR calibration
bias, we compared two estimation approaches: a simple mean bias that does not
take into account beam blockage, and a weighted mean bias that considers the
quality value of each sample as weights. The corresponding standard deviation
and weighted standard deviation were calculated as well. The overall process
is summarized in Fig.
Flowchart describing the processing steps to calculate the mean bias
and the weighted mean bias between ground radar data and satellite radar
data. The results of each step are shown in
Sect.
In order to promote transparency and reproducibility of this study, we mostly
followed the guidelines provided by
The entire processing workflow is based on wradlib
Reading the TRMM 2A23 and 2A25 version 7 data, GPM 2AKu version 5A data, and
Subic ground radar data in the netCDF format converted through the EDGE
software of EEC radars was done through the input–output module of wradlib.
The beam blockage modelling is based on the
An accompanying GitHub repository that hosts the Jupyter notebooks of the
workflow and sample data is made available at
From the 183 TRMM and 103 GPM overpasses that intersected with the 120 km
Subic radar range, only 74 TRMM and 40 GPM overpasses were considered valid
after applying the selection criteria listed in
Table
For the TRMM overpass event on 8 November 2013, the top row of
Fig.
GR-centered maps of volume-matched samples from 8 November 2013 at a
0.5
Consequently, the estimate of the calibration bias substantially depends on
the consideration of partial beam blockage (or quality). Ignoring quality
(simple mean) yields a bias estimate of
This case demonstrates how partial beam blockage affects the estimation of GR
calibration bias. At a low elevation angle, substantial parts of the sweep
are affected by total beam blockage. The affected bins are either below the
detection limit, or they do not exceed the GR threshold specified in
Table
Same as in Fig.
The effect becomes obvious for the next elevation angle.
Figure
The second case confirms the findings in the previous section for a GPM
overpass on 1 October 2015. That overpass captured an event in the northern
and eastern parts of the radar coverage where partial beam blockage is
dominant, as well as a small part of the southern sector with partial and
total beam blockage. Figure
As in Fig.
Finally, we applied both the simple and quality-weighted mean bias
estimations to each of the TRMM and GPM overpasses from 2012 to 2016 that met
the criteria specified in Sect.
As a result, we obtain a time series of bias estimates for GR calibration, as
shown in Fig.
The time series provide several important insights.
(1)
(2)
(3)
(4)
In 2011, Schwaller and Morris presented a new technique to match spaceborne radar (SR) and ground-based radar (GR) reflectivity observations, with the aim to determine the GR calibration bias. Our study extends that technique by an approach that takes into account the quality of the ground radar observations. Each GR bin was assigned a quality index between 0 and 1, which was used to assign a quality value to each matched volume of SR and GR observations. For any sample of matched volumes (e.g. all matched volumes of one overpass, or a combination of multiple overpasses), the calibration bias can then be computed as a quality-weighted average of the differences between GR and SR reflectivity in all samples. We exemplified that approach by applying a GR data quality index based on the beam blockage fraction, and we demonstrated the added value for both TRMM and GPM overpasses over the 115 km range of the Subic S-band radar in the Philippines for a 5-year period.
Although the variability of the calibration bias estimates between overpasses
is high, we showed that taking into account partial beam blockage leads to
more consistent and more precise estimates of GR calibration bias. Analyzing
5 years of archived data from the Subic S-band radar (2012–2016), we also
demonstrated that the calibration standard of the Subic radar substantially
improved over the years, from bias levels of around
Considering the scatter between SR and GR reflectivity in the matched volumes
of one overpass (see case studies), as well as the variability of bias
estimates between satellite overpasses (see time series), it is obvious that
we do not yet account for various sources of uncertainties. Also, the
simulation of beam blockage itself might still be prone to errors.
Nevertheless, the idea of the quality-weighted estimation of calibration bias
presents a consistent framework that allows for the integration of any
quality variables that are considered important in a specific environment or
setting. For example, if we consider C-band instead of S-band radars,
path-integrated attenuation needs to be taken into account for the ground
radar, and wet radome attenuation probably as well
In addition, with the significant effort devoted to weather radar data
quality characterization in Europe
Despite the fact that there is still ample room for improvement, our tool
that combines SR–GR volume matching and quality-weighted bias estimation is
readily available for application or further scrutiny. In fact, our analysis
is the first of its kind that is entirely based on open-source software, and
is thus fully transparent, reproducible and adjustable (see also
Through these open-source resources, our methodology provides both research institutions and weather services with a valuable tool that can be applied to monitor radar calibration, and – perhaps more importantly – to quantify the calibration bias for long time series of archived radar observations, basically beginning with the availability of TRMM radar observations in December 1997.
Code and sample data can be accessed at
IC and MH conceptualized the study. KM, MH, RW, and IC formulated the 3D-matching code based on previous work of RW. IC carried out the analyses; IC and MH the interpretation of results. IC and MH, with contributions from all authors, prepared the manuscript.
The authors declare that they have no conflict of interest.
The radar data for this analysis were provided by the Philippine Atmospheric,
Geophysical and Astronomical Services Administration (PAGASA,