Advantages of GSMaP Data for Multi‐Timescale Precipitation Estimation in Luzon

The Global Precipitation Measurement Mission (GPM) provides two different sources of post‐real‐time satellite‐based rainfall estimates, including the Integrated Multi‐satellitE Retrievals for GPM Final Run (herein IMERG‐F) and Global Satellite Mapping of Precipitation‐Gauge (herein GSMaP‐G). However, relative to IMERG‐F, GSMaP‐G has been less thoroughly evaluated in the context of studying rainfall variations over Luzon, Philippines. Using rain‐gauge observations over Luzon as a reference base, this study aimed to clarify if GSMaP‐G v07 is more capable than IMERG‐F v06 in regard to representing rainfall variations over Luzon at multiple timescales (including intraseasonal, annual, and interannual). The results revealed that both IMERG‐F and GSMaP‐G performed better in regard to depicting the temporal phase evolution for the wet season rainfall than they did for the dry season rainfall for Luzon. However, relative to IMERG‐F, GSMaP‐G exhibited better performance in most examined features, including (a) the detection of temporal variations (both phase and amplitude) of rainfall on the intraseasonal, annual, and interannual timescales; (b) the representation of spatial differences with more rainfall occurring from April to September for western Luzon and from October to March for eastern Luzon; (c) the detection of occurrence of rainfall events at various intensities. The analysis also demonstrates that the difference between the performance of IMERG‐F and GSMaP‐G in regard to quantitative rainfall estimation is larger during the wet season than it is during the dry season. The possible cause for their performance differences over Luzon is the differences in the gauge‐adjustment method and the biases in its pre‐non‐gauge‐adjusted product.

. Therefore, in addition to the global perspective, it is important to compare the performance of SREs from a regional perspective.
Luzon is characterized by complex topography and is the largest island in the Philippines and in the South China Sea region. Rainfall variation over Luzon is known to be modulated by multiple timescale atmospheric circulation changes, including seasonal and interannual timescale changes (Chang et al., 2005;Lee et al., 2021;Matsumoto et al., 2020). The seasonal variation in rainfall distribution over Luzon is affected by the northeast monsoon in the winter (Olaguera et al., 2018) and by the southwest monsoon in the summer (Akasaka et al., 2007;. The interannual variability of rainfall over Luzon is modulated by the occurrence of the El Niño-Southern Oscillation (Corporal-Lodangco et al., 2016). Due to its location, Luzon is also frequently affected by extreme rainfall in conjunction with active tropical convection (Bagtasa, 2017;Teng et al., 2021). These climate and weather characteristics make Luzon a special place for verifying the accuracy of SREs in regard to depicting multiple timescale rainfall variations.
Numerous studies Peralta et al., 2020;Sunilkumar et al., 2019;Veloria et al., 2021) have been conducted to evaluate the accuracy of GPM products over Luzon and the nearby regions. Among these studies, Sunilkumar et al. (2019) compared the performance of Tropical Rainfall Measuring Mission (TRMM) 3B42 v7 and IMERG-F v05 during 2014-2015, and they determined that IMERG-F can represent rainfall measurements better than can TRMM 3B42 over Luzon and the nearby regions. The ability of IMERG-F v06 to depict the monthly temporal variation of rainfall in the Philippines during 2014-2017 was also noted by Veloria et al. (2021). However, compared to the widely examined IMERG post-real-time product (i.e., IMERG-F), the performance of the GSMaP post-real-time product (i.e., GSMaP-G) in regard to capturing rainfall variations over Luzon is less discussed and has not been compared in detail to IMERG-F. Although the performance of non-gauge-adjusted GSMaP-M in regard to estimating the rainfall over Luzon has been examined by Aryastana et al. (2022), their findings cannot be directly applied for assessing the performance of gauge-adjusted GSMaP-G. It remains unclear which GPM-era post-real-time data (IMERG-F or GSMaP-G) is more suitable for capturing rainfall characteristics over Luzon at multiple timescales.
The primary objective of this study is to clarify this issue, as it is important for future applications of GPM-era post-real-time products in regard to studying rainfall variation over Luzon. We primarily examined the performance at intraseasonal, annual, and interannual timescales, and our goal was to understand if the then-current GPM-era GSMaP-G v07 is more suitable than is the IMERG-F v06 for use in depicting the related rainfall variations. The remainder of this manuscript is organized as follows. Section 2 introduces the data and analysis methods used. Section 3.1 presents the evaluation results for annual and interannual timescales, and the results for intraseasonal timescales are provided in Section 3.2. Section 3.3 provides a discussion of the possible explanations for the performance differences between IMERG-F and GSMaP-G, and this is followed by a conclusion in Section 4.

Data
The examination used the then-current versions of IMERG-F (i.e., v06) (Huffman et al., 2020) and GSMaP-G (i.e., v07) Mega et al., 2019). IMERG v06 was developed primarily by the United States National Aeronautics and Space Administration using the 2017 version of the Goddard Profiling Algorithm (GPROF2017) and is vital for tall-precipitation systems (Huffman et al., 2017). IMERG-F has provided rainfall data with temporal and spatial resolutions of 30 min and 0.1°, respectively, since 2000 (accessible from https://pmm.nasa.gov/data-access/downloads/gpm). On the other hand, GSMaP v07 was developed by the Japan Aerospace Exploration Agency, employing the passive microwave (PMW) precipitation retrieval algorithm, PMW-infrared combined algorithm, and gauge-adjustment algorithm. Notably, this version stands out for its capacity to accurately represent both "warm rain" and "orographic rain," which are distinctive features of rainfall in Asian monsoon . GSMaP-G provided rainfall data from March 2014 with temporal and spatial resolutions of 1 hr and 0.1°, respectively (accessible from https://sharaku.eorc.jaxa.jp/GSMaP/). For comparison, we accumulated the 30 min IMERG-F product to generate 1 hr rainfall data. The analysis focused on the overlapping time periods for the IMERG-F and GSMaP-G from March 2014 to February 2021. The evaluation used rain gauge observations (hereafter referred to RGO) that were obtained from the Global Surface Summary of the Day (GSOD) (Sparks et al., 2019) data (accessible from http://www1.ncdc.noaa.gov/pub/data/gsod). Due to some missing values or technical breakdowns, only data from 18 stations (hereafter STN 1-STN 18) can be used for the analyzed period from March 2014-February 2021. The full names and locations (longitude, latitude, and altitude) of the stations are listed in Table 1.

Methods
For the evaluation, we used the statistical methods provided in Equations 1 and 2 to compare the differences between the SREs and RGO.
Root mean square error (RMSE) = where SREi (RGOi) is a specific sample of SRE (RGO), N is the size of the sample, and SRE Additionally, we used the categorical metric (Table 2) to help identify samples as "Hits" "False alarms," and "Misses." Here, "Hits" indicates that SRE accurately detected the rainy events when RGO occurred (the definition of rainy events is explained later in Section 3.2). "False alarms" indicates that SRE overestimated the rainy events when RGO occurred. "Misses" indicates that SRE cannot capture the occurrence of rainy events when RGO occurred. These values were then used to calculate the three skill scores:

Probability of Detection (POD) = Hits Misses + Hits
(3) False Alarm Ratio (FAR) = False alarms False alarms + Hits (4) Critical Success Index (CSI) = Hits False alarms + Misses + Hits In Equations 3-5, POD, FAR, and CSI are the ratio of correctly estimated events, ratio of false alarm events, and ratio of correct diagnosis of events as detected by SRE, respectively (Ebert, 2007). The best values for TSCC, POD, and CSI were 1, and the best values for RMSE and FAR were 0. Hereafter, unless noted otherwise, the climatological mean is averaged over the analyzed periods from March 2014-February 2021. Figure 1a indicates the study area of Luzon and the horizontal locations of 18 stations. These stations can be grouped into two regions: western Luzon (STN 1-11) and eastern Luzon (STN 12-18). As presented in Figure 1a, most stations are located in coastal areas, and only a few stations are located in mountainous and inland areas. By comparing the altitudes of the stations (bars in Figure 1b) and the annual mean rainfall estimated by RGO (black line in Figure 1b), we noted that there is no linear relationship between these two variables; however, there is an obvious difference in the annual mean rainfall captured by RGO. By further comparing the stationary difference in annual mean rainfall observed by RGO to those estimated by GSMaP-G (blue line in Figure 1b) and IMERG-F (red line), we determined that GSMaP-G is more capable than is IMERG-F in regard to quantitatively depicting the regional differences as observed from RGO. Unlike RGO and GSMaP-G, IMERG-F detected fewer differences in the annual mean rainfall amount among the 18 stations. Statistically, the correlation coefficient between GSMaP-G and RGO in Figure 1b was approximately 0.94, and this was much higher than was that between IMERG-F and RGO (correlation coefficient = 0.52).

Analysis of Monthly Rainfall Variations on Annual and Interannual Timescales
The ability of the two SREs to capture the annual evolution of monthly rainfall variation was further examined. As presented in Figure 2a, RGO reveals that the annual evolution of monthly rainfall consists of a clear eastwest contrast. Overall, from April to September (AMJJAS) is the wet half-year for western Luzon (STN 1-11).
In contrast, from October to March (ONDJFM) is the wet half-year for eastern Luzon (STN 12-18). Among the stations, STN 3 and 5 exhibited larger annual variability than did the other stations. By comparing these features to those captured by IMERG-F ( Figure 2b) and GSMaP-G ( Figure 2c), we noted that both SREs are capable of depicting the east-west contrast observed in the annual rainfall variation. Using the time series of SREs (Figures 2b and 2c) and RGO (Figure 2a), we further computed the related TSCC for each of the 18 stations. As presented in Figure 2d, both SREs possess TSCC value >0.6 (marked by lines), and these are significant at the 95% confidence interval; this is true for all 18 stations. The difference between the TSCC in GSMaP-G and IMERG-F is that the former (blue line in Figure 2d) is more stable among stations, while the latter (red line) is closer to the perfect score for western Luzon than it is for eastern Luzon. As a result, GSMaP-G is clearly better than is IMERG-F in terms of the performance of the TSCC for eastern Luzon on an annual timescale. We also calculated the RMSE between the SREs (Figures 2b and 2c) and RGO ( Figure 2a) on an annual timescale, and the result is presented in Figure 2d (bars). We noted that GSMaP-G (blue bar in Figure 2d) exhibited a smaller RMSE value than did IMERG-F (red bar) for most stations. By subtracting IMERG-F from RGO, Figure 2e further indicates that IMERG-F tends to underestimate rainfall in the wet season (AMJJAS for western Luzon and ONDJFM for eastern Luzon) but overestimates rainfall in the dry season. In contrast, when compared to Note. "Yes" indicates that the rainfall events at a selected criterion are successfully captured, while the opposite situation is denoted as "No". RGO, GSMaP-G tended to overestimate rainfall for most stations regardless of the season ( Figure 2f). Relative to IMERG-F, GSMaP-G exhibits fewer errors in the quantitative estimation of rainfall variation on an annual timescale.
Notably, as there is a clear east-west seasonal difference that can be observed in Figure 2, we divided the annual cycle into two half-years (AMJJAS and ONDJFM) to evaluate the spatial rainfall characteristics. By doing this, we can better illustrate the seasonal and regional differences in the performance of SREs. The related spatial rainfall distributions are presented in Figure 3 for RGO (Figures 3a and 3d), IMERG-F (Figures 3b and 3e), and GSMaP-G (Figures 3c and 3f). Focusing on RGO, it is noted that during AMJJAS, most of the stations in western Luzon experienced more rainfall than did those in eastern Luzon (see circles in Figure 3a). Conversely, unlike AMJJAS, most of the stations in eastern Luzon experienced more rainfall than did those in western Luzon during ONDJFM (see circles in Figure 3d). Earlier study  has suggested that these east-west contrasts occurred in response to the prevailing winds (see vectors in Figures 3a and 3d) interacting with the local topography in Luzon. Consistent with RGO, both IMERG-F (Figures 3b and 3e) and GSMaP-G (Figures 3c and 3f) are capable of depicting the east-west contrast of season mean rainfall observed in AMJJAS and ONDJFM. However, relative to IMERG-F, GSMaP-G detected a larger east-west gradient in both the AMJJAS and ONDJFM mean rainfall. These results are consistent with Figures 2a-2c, indicating that GSMaP-G exhibits a better ability than IMERG-F in quantitatively depicting the east-west contrast of AMJJAS and ONDJFM mean rainfall over Luzon.
The climatological features are presented in Figures 2 and 3. To further clarify if the better ability of GSMaP-G in regard to depicting seasonal and regional differences in rainfall is a common feature observed for all years, we examined the interannual variation of AMJJAS ( Figure 4) and ONDJFM ( Figure 5) mean rainfall for 18 stations from 2014 to 2021. Overall, the RGO (Figures 4a and 5a) reveals that the east dry-west wet feature was present during all examined years of AMJJAS, in contrast to the west dry-east wet feature seen in ONDJFM. Among the stations, larger AMJJAS mean rainfall amounts were detected in STN 3 and STN 5 for 2015-2016 and By comparing these AMJJAS mean rainfall features to those of IMERG-F ( Figure 4b) and GSMaP-G (Figure 4c), we noted that both SREs also captured features similar to those of RGO ( Figure 4a); however, IMERG-F appeared to underestimate more than did GSMaP-G when depicting the maximum values of STN 3 and STN5 for 2018-2019 (see also Figures 4e and 4f). Further analysis of the related TSCC (see lines in Figure 4d; all pass the 95% significance test) and RMSE (bars in Figure 4d) for the interannual variation of AMJJAS mean rainfall revealed that GSMaP-G is better overall compared to IMERG-F. In particular, GSMaP-G exhibits a clearly higher value for TSCC compared to that of IMERG-F over eastern Luzon and a lower RMSE for most stations. Regarding the ONDJFM period, it is observed that both IMERG-F ( Figure 5b) and GSMaP-G ( Figure 5c) capture similar interannual features as RGO (Figure 5a). However, IMERG-F tends to underestimate rainfall over eastern Luzon (Figure 5e) to a greater degree than GSMaP-G (Figure 5f). This regional difference in performance between GSMaP-G and IMERG-F is further highlighted in the related statistical analysis provided in Figure 5d. The analysis shows that IMERG-F (red bar) possesses an RMSE larger than that of GSMaP-G (blue bar) mainly over eastern Luzon. Interestingly, despite the difference in magnitude observed in RMSE, both SREs exhibit similar performances in regard to TSCC for most stations (lines in Figure 5d; all pass the 95% significance test).
Thus far, we can briefly summarize from Figures 1-5 that GSMaP-G is more suitable than is IMERG-F in regard to depicting spatiotemporal rainfall variations over Luzon on annual and interannual timescales. Regarding the possible factors contributing to the interannual variation of AMJJAS and ONDJFM rainfall over Luzon, we constructed examinations to clarify whether it is determined by the interannual variation of tropical cyclones (TCs). The related results are presented in Supplementary Figures S1-S3 in Supporting Information S1, where we demonstrate the following findings: (a) Luzon is impacted by TCs in both AMJJAS and ONDJFM (refer to    Figure S1 in Supporting Information S1), (b) the number of TC cases and TC-related days affecting rainfall in Luzon accounts for less than 10% of the total days in AMJJAS and ONDJFM (refer to Figure S1 in Supporting Information S1), (c) the interannual features depicted in Figures 4a and 5a are primarily influenced by the non-TC events than the TC events (refer to Figures S2 and S3 in Supporting Information S1). For a more comprehensive analysis of the roles of TC and non-TC events in affecting the rainfall formation over Luzon, which falls beyond the scope of this study, we recommend referring to earlier literature (Akasaka et al., 2007;Bagtasa, 2017;Kubota & Wang, 2009;Matsumoto et al., 2020).

Analysis of Pentad Rainfall Variations on the Intraseasonal Timescale
Subsequently, we compared the east-west contrast of pentad (5-day average) rainfall as estimated by RGO, IMERG-F, and GSMaP-G for western Luzon (Figure 6a) and eastern Luzon (Figure 6b). The results shown in Figures 6a and 6b were averaged from March 2014 to February 2021. It is noted that in addition to the annual cycle, there are intraseasonal oscillation signals embedded in Figures 6a and 6b. Visually, it seems that both IMERG-F (red line in Figures 6a and 6b) and GSMaP-G (blue line) can capture intraseasonal oscillation signals similar to those of RGO (gray bar). However, it is not clear if the oscillation periods for western Luzon are similar to those for eastern Luzon, and the ability of SREs to depict these oscillation periods also remains unclear.
To clarify these questions, the oscillation period of pentad rainfall variations in Figures 6a and 6b was further explored using power spectrum analysis, and the results are presented in Figures 6c and 6d, respectively. Overall, RGO indicates that the most obvious oscillation period is approximately 20-40 days (with a maximum peak at 30 days) for western Luzon (black line in Figure 6c), while it is approximately 20-30 days (with a maximum peak at approximately 25 days) for eastern Luzon (black line in Figure 6d). Among the two SREs, GSMaP-G (blue line in Figures 6c and 6d) exhibit a closer performance than IMERG-F (red line) to RGO (black line) in capturing the intraseasonal oscillation period for both western and eastern Luzon. It is also interesting to note that the results of GSMaP-G and IMERG-F are more similar to each other for Figure 6c than it is for Figure 6d, thus implying that GSMaP-G and IMERG-F may exhibit a more similar performance in TSCC for illustrating the pentad rainfall variation over western Luzon than they do for eastern Luzon.
To clarify the above implication, we constructed scatter plots of the pentad rainfall as estimated by the two SREs versus RGO and divided the results for eastern and western Luzon. We also separated the results into two seasons that included AMJJAS ( Figure 7) and ONDJFM ( Figure 8). Indeed, consistent with what is inferred from Figures 6c and 6d, Figures 7 and 8 indicate that the two SREs possess a closer TSCC value (given in the top left corner) for western Luzon than they do for eastern Luzon in regard to the pentad data variation, and this is true for both AMJJAS and ONDJFM. Focusing on AMJJAS, GSMaP-G (Figures 7c  and 7d) is closer to the 1:1 fit line than is IMERG-F (Figures 7a and 7b) for both western and eastern Luzon, ultimately leading to a smaller RMSE value (left top corner) in GSMaP-G than that observed in IMERG-F when compared to RGO. As a result of the higher TSCC and lower RMSE, we suggest that GSMaP-G outperformed IMERG-F in regard to depicting the pentad rainfall variation over Luzon during AMJJAS. It should be noted that similar performance differences between GSMaP-G and IMERG-F are also revealed in Figure 8, showing that GSMaP-G (Figures 8c and 8d) outperformed IMERG-F (Figures 8a and 8b) with a scatter distribution of pentad rainfall variation closer to the 1:1 fit line. As a result, a higher TSCC and a lower RMSE were revealed in GSMaP-G compared to those for IMERG-F during ONDJFM, and this is true for both western Luzon and eastern Luzon.
Using the sample sizes presented in Figures 7 and 8, we further calculated the skill scores (POD, FAR, and CSI) for the pentad rainfall variation at each of the 18 stations. Results are presented as boxplots for western and eastern Luzon and for AMJJAS and ONDJFM in Figure 9. Here, the bottom, median, and top of the box represents the 25th, 50th, and 75th percentile of the focused skill score. The top and bottom lines or points extending out of the box represent the maximum and minimum values, respectively. The results indicate that for most samples over western Luzon, GSMaP-G exhibits a higher POD, CSI, and a lower FAR than does IMERG-F, and this is true for both AMJJAS (Figure 9a) and ONDJFM (Figure 9c). These performance differences imply that GSMaP-G possesses a better ability than does IMERG-F to detect the occurrence of pentad rainfall events over western Luzon. For eastern Luzon, GSMaP-G also exhibited POD and CSI values that were higher than those of IMERG-F during both AMJJAS ( Figure 9b) and ONDJFM (Figure 9d). However, IMERG-F and GSMaP-G outperformed each other in terms of FAR over eastern Luzon in ONDJFM and AMJJAS, respectively. Despite  this difference, most skill scores still suggested that GSMaP-G outperformed IMERG-F in regard to detecting the occurrence of pentad rainfall events over eastern Luzon.
Finally, we classified the pentad rainfall data from March 2014 to February 2021 into several criteria for rainfall intensities and calculated the related POD, CSI, and FAR ( Figure 10). Following Peralta et al. (2020) and Jamandre and Narisma (2013), we classified rainfall data into non-rainy events (0-1 mm·d −1 ), light-to-moderate rain events (1-10 mm·d −1 ), heavy rain events (10-20 mm·d −1 ), and extreme rain events (≥20 mm·d −1 ). The extreme rain events were further grouped into two ranges that included 20-50 mm·d −1 and ≥50 mm·d −1 . Here, we did not separate the results into AMJJAS and ONDJFM to obtain representative sample sizes for extreme rainfall. Overall, for all rainfall intensities examined GSMaP-G outperformed IMERG-F with a larger value of POD (Figure 10a), a larger value of CSI (Figure 10b), and a smaller value of FAR (Figure 10c). This indicates that our earlier suggestion of GSMaP-G possessing better performance in regard to depicting rainy events is less dependent upon the selection of rainfall intensity.
Furthermore, in Figure 10 we observed that POD and CSI exhibit their minimum values within the range of 10-20 mm·d −1 , while FAR reaches its maximum at the same rainfall intensity range. To clarify the potential cause behind these observations, we have included Table S1 in Supporting Information S1, which presents the number of samples for each rainfall intensity. In Table S1 in Supporting Information S1, the sample size of RGO represents the number of detected rainy events at different intensities during March 2014-February 2021. As for IMERG-F and GSMaP-G, the respective numbers in Table S1 in Supporting Information S1 represent the samples size that accurately detected the specific rainfall intensity event matching the RGO (i.e., the number of hits in Table 2). Upon examining Table S1 in Supporting Information S1, it appears that there is an inverse relationship between the rainfall intensity and the sample size not only in RGO but also in two SREs. Specifically, as the rainfall intensity increases, the number of accurately detected samples decreases. However, when compare the ratio between the sample sizes of IMERG-F and RGO in Table S1 in Supporting Information S1, we can observe a minimum value within the rainfall intensity range of 10-20 mm·d −1 . Similar feature is also noted in the ratio between the sample sizes of GSMaP-G and RGO. This indicates that the performance of IMERG-F and GSMaP-G in accurately detecting the rainfall intensity within the range of 10-20 mm·d −1 is inferior to other rainfall intensities. Other explaination regarding to the performance discrepancies of IMERG-F and GSMaP-G in detecting different rainfall intensities is further discussed in Section 3.3.

Discussion
One may question why GSMaP-G performed better than did IMERG-F in regard to detecting rainfall characteristics over Luzon. We believe that the likely reason is either (1) the differences in the gauge-adjustment method or (2) the differences in the performances of their pre-non-gauge-adjusted products (i.e., GSMaP-M and IMERG-L). For the first reason, it is known that IMERG-F uses the Global Precipitation Climatology Centre (GPCC), whereas GSMaP-G uses the NOAA Climate Prediction Center (CPC) gridded precipitation data for gauge adjust- ment. Therefore, if the gauge-adjustment method plays an important role in regard to affecting the performance difference between the IMERG-F and GSMaP-G over Luzon, we may observe a similar performance difference between the GPCC and CPC over Luzon.
To further investigate the possible reason for the performance differences between GPCC and CPC, we conducted additional analyses. We compiled information on the number of rain gauges used in the GPCC and CPC rainfall products (shaded regions in Figure S4 in Supporting Information S1) and compared them with the locations of the rain gauge observations (red dots with station numbers in Figure S4 in Supporting Information S1). As noted from Figures S4a and S4b in Supporting Information S1, the number of stations for each grid box in GPCC and CPC (shaded) matched well with the reference based-number of rain gauges (i.e., the number of red dots within the grid box). However, notable differences between GPCC and CPC can be observed in certain areas, particularly where there is a higher concentration of rain gauges. For example, in the region spanning 120°−121°E and 16°−17°N, GPCC covers STN 3 and STN 4 ( Figure S4a in Supporting Information S1), while CPC assigns STN 3 and STN 4 to different grid boxes ( Figure S4b in Supporting Information S1). This indicates that CPC captures more regional details compared to GPCC. Consequently, when utilizing the higher resolution of CPC for gauge-adjustment of satellite products, it is likely to yield results that are closer to the reference-based rain gauge observations.
Next, we compared the annual mean rainfall estimates (averaged from March 2014 to February 2021) obtained from the RGO (bars in Figure 11a), GPCC (red open circles in Figure 11a), and CPC (blue closed circles in  Figure 11a). It is worth noting that GPCC and CPC exhibit closer agreement with each other, with CPC demonstrating slightly better performance compared to GPCC for 10 stations (as shown in Table S2 in Supporting Information S1). This similarity between CPC and GPCC is not only observed in the area-averaged values at the aggregated annual scale but also holds true for each specific month (as illustrated in Figure S5 in Supporting Information S1). This finding is further supported by the correlation coefficient between CPC and GPCC (as presented in Table S3 in Supporting Information S1), with all values exceeding 0.8 and passing the 90% significance test. However, when comparing to the RGO in Figure 11a as a reference, CPC exhibits a higher TSCC (=0.89) and a lower RMSE (=1.04 mm·d −1 ) compared to those of GPCC (TSCC = 0.78 and RMSE = 1.27 mm·d −1 ). This provides evidence to support again that CPC is closer to RGO compared to GPCC. These results support our hypothesis that the performance differences between GSMaP-G ( Figure 11c) and IMERG-F (Figure 11b) may be partially attributed to the gauge-adjustment method. Notably, GSMaP has larger adjustments from the multi-satellite to the satellite-gauge than IMERG; this is not only revealed in the aggregated annual scale (Figures 11b and 11c) but also holds true for most months ( Figure S6, Table S4 in Supporting Information S1).
Regarding the possibility of the aforementioned second reason, a number of hints can be inferred from the results of earlier studies (Wang & Yong, 2020). Wang and Yong (2020) examined the non-gauge-adjusted products of the GSMaP-M and IMERG-L over global land areas. Although Wang and Yong (2020) reported that IMERG-L outperforms GSMaP-M over most global land areas, we note from Figure 1 that IMERG-L is clearly underestimated over the regions surrounding Luzon. The evidence supporting this finding is presented in Figure 11d that indicates that the errors of IMERG-F in regard to underestimating rainfall over Luzon originally existed in IMERG-L. Possible reasons for the underestimation of rainfall by IMERG-L have been suggested in earlier studies (Bogerd et al., 2021;Wang et al., 2021) that indicate that IMERG-L tends to miss shallow rainfall events, and this may lead to underestimation of rainfall. Additionally, we note from Figures 11d and 11e that GSMaP-M is better than is IMERG-L for assessing eastern Luzon but not for western Luzon. Comparing this finding to the results presented in Figure 1b, it is likely that in addition to the gauge-adjustment method, the better performance of GSMaP-M compared to that of IMERG-L also contributes to the better performance of GSMaP-G compared to that of IMERG-F for eastern Luzon. In contrast, for western Luzon the gauge-adjustment method may be the main reason for why GSMaP-G performs better than IMERG-F. These findings suggest that the reasons for the performance differences between the IMERG-F and GSMaP-G are also regionally dependent.
Furthermore, the algorithmic characteristics of IMERG and GSMaP, as well as their performance in representing the rainfall system over Luzon, also contribute to the observed differences. Previous studies suggest that Luzon is predominantly influenced by "orographic rain" (Hirose et al., 2017), and afternoon showers and mesoscale convective systems are the main extreme rainfall patterns in the inland areas of Luzon (Hamada et al., 2014). Considering that (a) "orographic rain" plays a crucial role in Luzon (Hamada et al., 2014;Hirose et al., 2017), and (b) GSMaP is known for its ability to capture "warm rain" and "orographic rain" in Asian monsoon rainfall , it can be inferred that these factors partially explain why GSMaP-G performs better than IMERG-F in depicting the variations in Luzon rainfall. However, further evidence is required to confirm this suggestion. One approach could involve utilizing, the Quality Assurance (QA) flag data available in GSMaP's near-real-time product to diagnostically and quantitatively evaluate the factors responsible for the accuracy differences between the two products (Yamaji et al., 2021). This analysis could involve examining differences in precipitation during microwave radiometer passage (i.e., microwave retrieval differences) and differences in rainfall during geostationary meteorological satellite observations (i.e., interpolation algorithm differences). Nevertheless, currently the GSMaP QA flag data is only available for the near-real-time products, and thus not been further examined in the current study.

Conclusions
This study examines the ability of the then-current GPM-era post-real-time products from IMERG-F v06 and GSMaP-G v07 to detect multiple timescale (intraseasonal, annual, and interannual) rainfall variations over Luzon. The analyses focused on the overlapping data periods from March 2014 to February 2021. Using 18 local rain gauge observations (i.e., RGO) as a reference base for performance assessment, we determined that rainfall variation over Luzon consists of clear seasonal and regional differences. The RGO indicates that AMJJAS and ONDJFM are wet half-years for western and eastern Luzon, respectively. Both IMERG-F and GSMaP-G are capable of depicting these east-west seasonal differences in a manner similar to that of RGO; however, IMERG-F exhibits a larger error in quantitative rainfall estimation with a clear underestimation of the monthly rainfall over the wet half year (AMJJAS for western Luzon and ONDJFM for eastern Luzon) (Figures 2 and 3). The analysis also indicated that both IMERG-F and GSMaP-G are capable of depicting the temporal phase of rainfall variation (with TSCC values passing the 95% significance test) on the annual (Figure 2), interannual (Figures 4 and 5), and intraseasonal (Figures 6-8) timescales. However, relative to IMERG-F, GSMaP-G tended to exhibit slightly better performance in regard to illustrating the temporal phase of rainfall variation (i.e., higher TSCC) for all examined timescales. Regarding the ability to detect the occurrence of rainy events at various intensities, our examinations of the skill scores indicate that GSMaP-G possesses a higher PDO, a higher CSI, and a lower FAR for AMJJAS and ONDJFM and for most examined stations (Figures 9 and 10), thus suggesting a better ability for GSMaP-G than for IMERG-F in depicting rainy events at various intensities.
Notably, although numerous studies have evaluated the performance of SREs over Luzon, this study is the first to examine the details in GSMaP-G and to compare its performance to that of the IMERG-F. To the best of our knowledge, this study is the first to indicate that the performance difference between the IMERG-F and GSMaP-G is larger in eastern Luzon than it is in western Luzon. In contrast to Huang et al. (2020) who reported that IMERG-F is better than GSMaP-G in regard to capturing rainfall variation over Taiwan (an island near Luzon), we demonstrated that GSMaP-G is better than is IMERG-F in regard to assessing most examined features over Luzon. This finding is important for further studies interested in using GPM post-real-time products to study rainfall variations in Luzon. However, as GSOD does not provide RGO on hourly timescales, we did not examine the ability of GSMaP-G and IMERG-F to capture diurnal rainfall variation over Luzon. This is suggested as a future research study. Furthermore, since our research was based on the IMERG v06 product, the answers to our research questions may potentially change with the release of the IMERG v07 product (Huffman et al., 2023). Therefore, we recommend conducting further examinations in future to evaluate the performance of IMERG v07 in depicting the rainfall variations over Luzon.