5 Discussion, Limitations, Outlook

Precipitation is the main driver of environmental and hydrological processes, and has a large impact in terms of natural hazards (through floods, debris flow, landslides, avalanches). Hence, there is an increasing need for more reliable precipitation estimates and forecasts at high temporal and spatial resolution. The recently-acquired Subic radar data allowed a first look at the capabilities and advantages it offers in weather monitoring in the Philippines. Radar data was used to reconstruct the enhanced southwest monsoon event of 2012 over Metro Manila. Torrential rainfall fell continuously over the course of four days, as the tail of Typhoon Haikui passing north of the Philippines pulled in even more moisture along with the already ongoing monsoon winds and rains. Radar-derived rainfall estimates underestimated by as much as 60% when compared with gauge measurements. Gauge-adjusted radar data reconstruction showed that while Metro Manila received the most rainfall over land, most of the rain actually fell over Manila Bay. This feature of the rainfall distribution would not have been identified from rain gauge interpolation alone, as demonstrated by the Supplemental figure of Chapter 2, as well as Figure 2a and c in a later study by (Abon et al. 2016). Additionally, the radar-based rainfall distribution map showed other localized areas (about 15–25 km wide) of high rainfall accumulation which were also missed in the rain gauge interpolation map (Supplemental figure of Chapter 2). These plume-like features re-appeared in the radar-based rainfall distribution map of a similar event (southwest monsoon enhanced by a north-passing typhoon) the following year, which were driven by the interaction of stratovolcanoes with the monsoon (Lagmay et al. 2015).

The archipelagic nature of the country’s geography prohibits a dense and well-spaced network of rain gauges that can be used to effectively compare gauge measurements and radar rainfall-estimates. The highly convective characteristic of local rainfall produces strong thunderstorms that can be compact enough to travel in between rain gauges and be left undetected. With the sparseness of the gauges and the uncertainty that comes with the rainfall-rainrate transformation, as well as the lack of auxiliary calibrators such as disdrometers and radar profilers, we instead turned to relative calibration. Reflectivities from ground-based radars (GR) are compared with reflectivities from spaceborne radars (SR), namely TRMM (for data from 2012–2014) and GPM (for data from 2014–2016). A more common approach in comparing SR and GR is by reprojecting and interpolating the surface rainfall estimates onto either a common 2D cartesian grid on the Earth’s surface or the volumetric rainfall estimates onto a common 3D cartesian grid. The 3D volume matching method of (Schwaller and Morris 2011) used in this paper avoids uncertainties that could be introduced by interpolation by considering only the volumes of the SR beam and GR beam that intersect. In addition, directly comparing the primary measured quantity reflectivity circumvents the need to convert to rain rate, which is another potential source of uncertainties if rain gauges are used as a reference.

By avoiding these known potential additional sources of errors, this thesis was able to focus on the sources of systematic errors that could influence the comparison between SR and GR. The consideration of data quality started with the beam blockage fraction (BBF) for Chapter 3 and included path-integrated attenuation (PIA) for Chapter 4. These quality factors were then used as weights in taking the weighted-average of the differences between the SR and GR reflectivities for a single coincidence of an SR overpass and a GR sweep. The weighted-averaging approach allows low quality data resulting from BBF and PIA, sources of uncertainty that are heterogeneous in space, to be filtered out for a more consistent estimate of calibration bias, which is homogeneous in space.

In Chapters 3 and 4, it was shown that introducing a data quality framework in the comparison increases the consistency of bias estimates between two datasets. In a case study discussed in Section 3.4, reflectivity difference maps between SR overpass and GR sweep indicate that significant variability (with absolute differences up to 10 dB) can be found at the edges of the blind sector, due to partial beam blockage. The locations of the large absolute differences are consistent with the locations of the low data quality. Assigning weights to beam blockage fraction (0 for the poorest data quality due to total beam blockage up to 1 for the best data quality without any topographic obstacles) in calculating the calibration bias reduces the scatter and consequently decreases the standard deviation of the differences between SR and GR reflectivity (Figure 3.6), which in turn increases the confidence in the bias estimation. The same reduction in standard deviation occurs as well in the vast majority of overpasses. The GR–GR comparison in Chapter 4 similarly demonstrates the benefits of considering BBF and PIA. The case study in Figure 4.4 shows that individually, using \(Q_{BBF}\) and \(Q_{PIA}\) as weights reduces the standard deviation of the differences between the reflectivities from the two ground radars, but considering both quality indices at the same time decreases the standard deviation by almost half, as opposed to considering all points equally. The combination of quality factors also dramatically narrows down the distribution of the reflectivity differences. These results demonstrate that BBF and PIA do not only affect Quantitative Precipitation Estimation (QPE), but also the comparability of two radars. The advantages of quality-weighting therefore applies to both SR–GR comparisons and GR–GR comparisons.

Being able to determine the calibration bias of one radar in Chapter 3 using the SR–GR matching approach and applying the method and correction to two overlapping radars in Chapter 4 tells us that SR and GR reflectivities are consistent enough (after filtering out low data quality) to allow for calibration bias estimates. This consistency presents the potential of calibrating multiple radars in a network against a single reference, using a uniform approach. In this study, this was demonstrated for at least two radars in the Philippine radar network. The increased agreement between the two ground radars after calibration demonstrates the feasibility of SR as a stable travelling reference. The method also allows for bias estimation at instantaneous points in time, as opposed to gauge calibration where errors are lumped over an hourly or daily timeframe. Due to its unobstructive nature, this relative calibration method can also be done continuously, so that calibration monitoring can be done even while the radars are operational.

Chapter 3 also discusses the consistency of TRMM and GPM with respect to each other. In the analysis of the bias time series for Subic radar, TRMM and GPM overpasses were both available for 2014. The period of overlap showed that GR calibration bias estimates based on both TRMM and GPM observations can be considered homogeneous. Based on this assessment, TRMM and GPM were lumped together as a continuous dataset when comparing SR and GR reflectivities for bias correction in Chapter 4.

It was discussed in the latter part of Chapter 3 (Section 3.4.2) that the bias fluctuates around an average value that varies year by year and appears to be quite persistent over the duration of the corresponding wet seasons from 2012–2016. The drastic change in bias between 2013 and 2014 for the Subic radar (Figure 3.8) has to be assumed to be due to calibration maintenance in terms of hardware changes (i.e. magnetron replacement). Unfortunately, similar information about the Tagaytay radar’s maintenance history was not available. Detailed information regarding maintenance protocols would be useful in explaining the changes in bias of radars throughout the years as demonstrated in Warren et al. (2018). Chapter 4 (Section 4.4.4.) reaffirms us that there is substantial short-term variability. Even after identifying and addressing the effects of BBF and PIA, heavy fluctuations of bias persist. The causes of these fluctuations could not yet be disentangled. Yet, we were able to reduce the mean absolute difference before and after bias correction when using interpolated bias estimates. So irrespective of the actual causes of fluctuation, the variability exhibits some level of continuity, so that an interpolation of bias estimates in time contributes to an overall improvement.

Aside from internal calibration of the precipitation radars onboard TRMM/GPM, external monitoring is also carried out through comparison with selected well-calibrated ground radars. Several ground validation sites can be found in different locations around the world (Hou et al. 2013). Some of these sites are established primarily for this purpose, while some are part of existing radar networks whose calibration are deemed accurate enough to use as reference. The geometry matching method of Schwaller and Morris (2011) that was used in Chapters 3 and 4 was developed with this application in mind—to compare reflectivities between ground radars and spaceborne radars to check if they are consistent with each other. While ground validation sites are typically set up to have the least possible errors, the method can still be applied for a more extensive global approach in ground validation using radars not included in the validation radars list. The consistency of the ground radar observations with the spaceborne radar observations in Chapter 3 also serves, albeit indirectly, as a validation of the SR estimates in the study region.

References

Abon, Catherine Cristobal, David Kneis, Irene Crisologo, Axel Bronstert, Carlos Primo Constantino David, and Maik Heistermann. 2016. “Evaluating the Potential of Radar-Based Rainfall Estimates for Streamflow and Flood Simulations in the Philippines.” Geomatics, Natural Hazards and Risk 7 (4): 1390–1405. https://doi.org/10.1080/19475705.2015.1058862.

Hou, Arthur Y., Ramesh K. Kakar, Steven Neeck, Ardeshir A. Azarbarzin, Christian D. Kummerow, Masahiro Kojima, Riko Oki, Kenji Nakamura, and Toshio Iguchi. 2013. “The Global Precipitation Measurement Mission.” Bulletin of the American Meteorological Society 95 (5): 701–22. https://doi.org/10.1175/BAMS-D-13-00164.1.

Lagmay, Alfredo Mahar F., Gerry Bagtasa, Irene A. Crisologo, Bernard Alan B. Racoma, and Carlos Primo C. David. 2015. “Volcanoes Magnify Metro Manila’s Southwest Monsoon Rains and Lethal Floods.” Frontiers in Earth Science 2: 36. https://doi.org/10.3389/feart.2014.00036.

Schwaller, Mathew R., and K. Robert Morris. 2011. “A Ground Validation Network for the Global Precipitation Measurement Mission.” Journal of Atmospheric and Oceanic Technology 28 (3): 301–19. https://doi.org/10.1175/2010JTECHA1403.1.

Warren, Robert A., Alain Protat, Steven T. Siems, Hamish A. Ramsay, Valentin Louf, Michael J. Manton, and Thomas A. Kane. 2018. “Calibrating Ground-Based Radars Against TRMM and GPM.” Journal of Atmospheric and Oceanic Technology, February. https://doi.org/10.1175/JTECH-D-17-0128.1.