Prediction of the Kamchatka July 29, 2025, earthquake by the evolution of low-magnitude seismicity recovered using waveform cross-correlation at IMS seismic arrays (Part 2)

 There are dozens of REB events in Figure 21 that are not matched in the XSEL. The reason behind this underperformance could be related to the WCC processing and the quality of the REB events obtained during the interactive review. Figure 22 displays the frequency distribution of SNR values for the first P-phases associated with all REB events on July 30, 2025. All SNR estimates are sourced from the IDC database. The pattern is controversial. There are 5241 arrivals with SNR>10, which are high-quality signals not shown in Figure 22 because they are well matched by the XSEL. The curve between 0 and 10 includes 15802 values in 0.2-wide bins. There are 4872 detections with SNR<3.4, with 244 of them having a default value of -1.0 and fixed at the SNR=0 point.  They fall below the automatic detection thresholds at the IMS stations and were added by IDC analysts. The peak of the distribution below 3.4 is around 2.0, with 2049 arrivals having SNR <2.0, and 321 with SNR <1.0. 

 

Figure 22. Frequency distribution of SNR values reported by the IDC for all P-phases associated with the REB events on July 30, 2025.

           

The quality of REB detections added by analysts is formally defined only by their experience and the task not to miss events of concern. The SNR estimates for the added signals, assuming they are correct, pose a challenge for any detector, including those based on the WCC. These detections were missed by the IDC automatic detector primarily due to their low quality. Furthermore, detections lacking SNR estimates should not be included in the database, as their quality is crucial for GA processing. The XSEL, when based on routine WCC processing, may encounter internal limitations in recovering events during periods of extremely high activity, which consequently affects the quality of the REB events.

 

Statistical power

The XSELs include numerous events with statistical significance controlled by the LA parameters for a given detection list. The question of their statistical power is related to the match of REB events in the XSELs with various LA versions and origin-time tolerance. The strict LA version with low tolerance has to be adjusted to create event hypotheses matching the REB and a small number of new XSEL events which could match the REB-potentially missed in the IDC processing and recovered by the WCC. Table 4 lists the match statistics for two LA versions and the origin-time tolerance of 2.0 s, as adopted by the REB.  The first number in the “XSEL matched” column is for the weak version and the second for the strict version. The “REB only” and “SEL3” columns split the “REB total” into automatic and interactive production stages. “XSEL LA matched” refers to the match by XSELs for individual MEs before the CR is applied.

 

Table 4. Statistics of REB and SEL3 match in the XSEL for the period 2025201-2025210.

Day of 2025	Q	REB total	XSEL matched	REB only	XSEL matched	SEL3	XSEL matched	XSEL LA matched	New XSEL
201	Q2	109	104 / 81	27	22 / 8	82	82 /73	108/96	113 / 15
	Q3	75	70 /63	14	12 / 8	61	59 / 55	74 / 69	120 / 17
	Q4	51	50 / 44	8	8 / 3	43	42 /41	51 / 48	113 /27
202	Q1	31	30 / 29	4	3 / 2	27	27 / 27	30 /30	120 / 24
	Q2	45	38 / 37	10	6 / 5	35	32 / 32	41 / 41	129 / 24
	Q3	44	42 / 41	10	9 / 5	34	33 / 33	44 /42	141 / 25
	Q4	23	22 / 18	5	4 / 1	18	18 / 17	22 / 18	121 / 23
203	Q1	55	52 / 46	20	17 / 13	35	35 / 33	55 / 50	102 / 16
	Q2	31	30 / 30	4	3 / 4	27	27 / 26	30 / 30	116 / 23
	Q3	20	20 / 18	8	8 / 6	12	12 / 12	20 / 18	118 / 24
	Q4	11	10 / 10	2	1/ 1	9	9 / 9	11 / 11	112 /18
204	Q1	4	4 / 4	0	0	4	4 / 4	4 / 4	101 / 24
	Q2	5	5 / 5	1	1 / 1	4	4 / 4	5 / 5	123 / 33
	Q3	11	11 / 11	0	0	11	11 / 11	11 /11	134 / 47
	Q4	9	9 / 9	2	2 / 2	7	7 / 7	9 / 9	91 / 19
205	Q1	7	7 / 7	1	1 / 1	6	6 / 6	7 / 7	99 / 24
	Q2	10	10 / 10	3	3 / 3	7	7 / 7	10 / 10	105 / 17
	Q3	6	6 / 5	3	3 /2	3	3 / 3	6 /6	133 / 37
	Q4	22	22 / 22	5	5 / 5	17	17 / 17	22 /22	131 / 28
206	Q1	16	16 / 16	4	4 / 4	12	12 / 12	16 / 16	164 / 45
	Q2	4	4 / 4	1	1 / 1	3	3 / 3	4 / 4	137 / 36
	Q3	2	2 / 2	0	0	2	2 / 2	2 / 2	129 / 33
	Q4	6	6 / 6	0	0	6	6 / 6	6 / 6	122 / 23
207	Q1	3	3 / 3	0	0	3	3 / 3	3 / 3	87 / 15
	Q2	4	4 / 4	1	1 / 1	3	3 / 3	4 / 4	107 / 28
	Q3	9	9 / 9	1	1 / 1	8	8 / 8	9 / 9	90 / 19
	Q4	4	4 / 4	0	0	4	4 / 4	4 / 4	107 / 37
208	Q1	0	0	0	0	0	0	0	110 / 24
	Q2	4	4 / 4	1	1 / 1	3	3 / 3	4 /4	89 / 16
	Q3	0	0	0	0	0	0	0	93 / 27
	Q4	5	5 / 4	2	2 / 1	3	3 / 3	5 / 5	89 / 24
209	Q1	3	2 / 3	2	1 / 2	1	1 /1	3 / 3	131 / 29
	Q2	2	2 / 2	0	0	2	2 / 2	2 /2	84 /  16
	Q3	1	1 / 1	0	0	1	1 / 1	1 /1	80 /  13
	Q4	1	1 / 1	0	0	1	1 / 1	1 / 1	135 / 38
210	Q1	2	2 /2	0	0	2	2 / 2	2 / 2	127 / 33
Total		635	604 / 555	139	119 / 81	496	472 / 454	626 /593	4103 / 921

 

The period between the J20 and J29 earthquakes is likely the best time to assess the statistical power of the XSEL events. Two days following the J20 are characterized by rates ranging from 100 to 30 REB events per 6 hours. The second quarter-day (Q2) of 2025201 includes 109 REB events. The XSEL for the weak LA version matched 104 out of 109, and the strict version matched 81. This difference between the weak and strict versions is relatively large, but it is decreasing with time from the J20. Since the second half of 2023203, there is no difference in the matches between the weak and strict versions for SEL3 events or for the pure REB. The watershed is likely 30 events per quarter-day. The larger the number of REB events the lower the match rate, and the reason for this deviation is discussed in the previous section.

The final column in Table 4 lists the number of new XSEL events. The weak LA version produces 114±19 events per 6 hours, whereas the strict version produces 26±9 events per 6 hours for the studied period. This difference is expected, given to the design of the detection threshold, which results in a decreasing number of XSEL events as the pair order number or the corresponding magnitude threshold increases. The number of new events in the strict LA version is slightly larger than 50% to 100% of the number of REB events with the expectation of approximate equality, as related to the exercises with interactive review of XSEL bulletins. For the 0.25-second tolerance case, the number of XSEL events after the 2025203 ranges between 2 and 11. This pair of the LA version and tolerance case is designated as the transition to the REB. The statistical power of the XSELs used in this study is demonstrated by a perfect alignment with the REB, encompassing both actual and potential REB-ready events.

The importance of the results in Table 4 lies in the equal statistical significance of XSEL events that match the REB and those that do not. This implies that a complete match of the REB events during the period of relatively high activity between July 22 and July 29, 2025, can be extended to other XSEL events, thereby assigning them the same quality as the REB.  In turn, the XSELs for the period between 201 and 210 of 2025 were obtained using the same procedure as before the J20 earthquake. The WCC processing introduced in [Kitov, 2026] and further developed in this study can be applied to IMS data for earthquakes across various regions. This approach guarantees both statistical significance and statistical power for the XSEL event hypotheses.

 

Results

            General aspects of WCC processing and its specific characteristics across different levels of seismic activity have been discussed. The seismicity within the studied region before and after the J20 and J29 earthquakes was also examined. The selection procedure and structure of the ME set were customized to fulfill the processing objectives. All calculations in this study were performed using identical WCC settings, ensuring direct compatibility throughout the entire processing period and across specific areas. The results include XSELs for individual MEs as well as the final XSELs obtained via the CR process. The full processing period spans from July 12 to July 31. Additionally, the best 100 MEs were subjected to an extended processing period from July 6 to August 3. Finally, two 3-day periods without REB events were processed using this 100 best ME set, providing a baseline reference for XSEL interpretation.

 

XSEL seismicity during low seismicity and periods without REB events

            The two 3-day periods without REB events allow for the estimation of recurrence curves exclusively for XSEL events. The first period, spanning November 29 to December 1, 2023 (jdate = 2023333-2023335), can be extended into the second quarter (Q2) of day 2023332. Additional calculations were performed for December 2 (5 REB events) and December 3 (1 event) to analyze the transition from a quiet period to the activity captured in the REB events. Similarly, the quiet period from February 24 (2024055) to February 26 (2024057), 2024, was also extended by two days exhibiting REB activity. Because this study focuses on rapid fluctuations in seismic activity below the IDC detection level, the data are processed in quarter-days intervals (Q1-Q4). Given that the activity levels can be exceptionally low, with zero XSEL events across several consecutive quarters, a four-quarter moving sum, denoted as MS(4), is calculated with a one-quarter step to ensure robust event statistics.

            The results of the WCC processing using the best 100 MEs are displayed in Figure 23. The upper panel shows the recurrence curves, where the x-axis represents the order numbers. These XSEL recurrence curves demonstrate the feasibility of converting these order numbers into magnitudes; furthermore, deviations from the regression lines are potentially related to fluctuations in seismic activity near a specific magnitude level. The curves between 333.125 (Q1/333) and 335.875 (Q4/335) follow exponential trends, exhibiting deviations from these trends that increase in amplitude as the order number increases. The summary curve for the entire 15-quarter quiet period yields an R²=0.98. On day 2023336, 5 REB events occurred, and two curves 336.875 and 337.125 – show a dramatic 3- to 4-fold increase in seismic activity throughout all the curves. This predominant growth is concentrated within the mid-range tolerance values of the strict LA version, peaking between 1.0 s and 2.0 s. A similar behavior is observed in the curves for some, but not all, quiet days.  The most prominent deviations occur toward the end of the quiet period, while the summary curve shows a low-amplitude deviation for a 1-second tolerance.

 

Figure 23. The recurrence curves of the moving sum of four consecutive 6-hours intervals for 12 XSELs from v₁c₁ to v₂c₆ with order number from 1 to 12.  Upper panel: days between Q2/2023332 to Q4/2023337. Each curve is marked by its end quarter:  curve 333.175 is the sum of 4 quarter-days between Q2/2023332 and Q1/2023333. Lower panel: the days between 2024055 and 2024059.

 

            The recurrence curves illustrate the integral characteristics of the XSEL event distribution over a linear scale, such as a magnitude-like one. This representation is important as it proves the quality of the observational tool and the unambiguous conversion of the order number of the version/case into a magnitude scale. The scale is not standard in terms of the measurement procedure underlying the values. There are many magnitude scales, each with slightly different characteristics, taken from different parts of the signal wavetrain from the first P-wave to Love waves. An inverse representation of these scales is also possible similar to that described for Figure 18.

The time evolution of the seismic process for a set of magnitude thresholds is crucial for understanding the earthquake preparation process. The numbers of XSEL events for 12 thresholds are shown in Figure 24 for the quiet period 2023332-2023335 with an extension into the days 2023336 and 2023337 with REB events. The XSEL curves, which represent the MS(4) of the previous four quarter-days, look smooth and synchronized for each LA version over the entire period including the quiet and active seismicity periods.

 

Figure 24. The evolution of the number of XSEL events in 6 weak cases (upper panel) and 6 strict cases (lower panel). Vertical dashed line – the origin time of the m_b=3.95 event occurred at 17:22:35 on December 2, 2023. Hypocenter is ~100 km due south of the J29 earthquake: 51.60ºN, 159.86ºE, d=0 km.

 

The origin time of the largest event on 2023336 with m_b=3.95 is indicated by a vertical red line. There is a noticeable increase in the XSEL numbers starting two quarter-days before this event. The only deviation is the drop in the 0.25 s curve for the strict LA version, which began approximately 20 hours before the earthquake. The ratios of the weak and strict LA versions for all six cases are displayed in Figure 25. These curves highlight the variances in seismic activity above the respective thresholds. The 0.25 s ratio curve begins to  rise three quarter-days before the earthquake, peaks roughly 8 hours prior to the earthquake, and then declines as it approaches the red line. A similar pattern was observed in the 0.25 s curve for the May 24, 2013, Sea of Okhotsk earthquake, while the other five curves exhibit different behaviors. The resemblance in the peaks in the 0.25 s curve is likely coincidental, but similar patterns before other major earthquakes are the objects of this study. For the 2023336 earthquake, the increase in the 0.25 s curve may be observable due to the absence of larger-magnitude seismic activity during the previous 96 hours.

 

Figure 25. The ratios of the XSEL for the weak and strict LA versions and the same origin time tolerance case. The curve for the 0.25 s tolerance has a peak approximately 10 hours before and then falls just before the earthquake at 17:22:25. This peak corresponds to the fall in the XSEL for strict LA version in Figure 23.

 

            The 3-day quiet period in 2024 is shown in Figure 26. It displays a pattern similar to that observed in 2023, with a peak in the 0.25 s ratio curve. Figure 27 depicts this curve starting to rise 5 quarter-days before peaking a day before the event on February 28 at 11:11:31. The epicenter coordinates are 53.93°N, 160.34°E, with a depth of 75.4 km, and m_b=3.85. This peak is observed long before the event origin time, and the event itself is not large. Therefore, there may be no causal link between the peak and the earthquake. Nevertheless, the peak and the events are measured independently and are where they are.

There were no significant changes observed in the XSEL numbers throughout the entire period between February 24 and February 29. However, the strict LA version demonstrated larger variations. The day of February 29 was added to the initial interval due to a sudden rise in the 0.25 s curve right up to the end of February 28, as observed in Figure 27. This rise is related to the decreasing 0.25 s curve for the strict LA version in the lower panel. The only REB event on February 29 occurred at 19:37:35 at a depth of 499 km, far to the west of the J29 epicenter (51.45°N, 151.28°E). The magnitude was 3.47, large enough for the event to be detected by almost all IMS stations worldwide.

igure 26. Same as in Figure 24 for the quiet period between 2024055 and 2024057 extended into 2024058-061. 

Figure 27. Same as in Figure 25 for the quiet period between 2024055 and 2024057 extended into 2024058-059.

 

Seismic activity prior to the J20 earthquake

            During the quiet periods, the seismic activity detected by the WCC is nearly identical to that observed during the periods of low activity periods preceding the J20 earthquake. Figure 28 displays the trajectories of the running four-quarter moving sum, MS(4), of the quarter-day-long XSELs between July 6 and July 20, 2025. This period was fully processed using the best 100 MEs and presents a baseline reference for relatively low seismicity prior to the J20 M7.4 earthquake (see Figure 19), which featured only one large REB event on July 14 (origin time: 21:38:01, epicenter: 46.66°N, 151.30°E, depth: 87 km, Mw~5.7-5.8). Results for both the weak (upper panel) and strict (lower panel) LA versions, each including six origin-time tolerance cases from 5.0 s to 0.25 s, are shown.

            Several intervals with significant variations in the MS(4) are marked by ovals. Between July 10 and 12, there was deep through, with all curves for both the weak and strict version changing almost synchronously. The strict/0.25 s curve dropped down to 1 XSEL per day during three consecutive quarter-days. There were four quarter-days without XSEL events in a row and only one XSEL event during a day and a half. There were also no REB events on 2025189. Given that the final REB event on 2025188 occurred at approximately 16:00 and the first event on 2025191 appeared at 06:00, the total event-free period lasted roughly 38 hours. The strict LA version with a 0.25 s origin time tolerance exhibits approximately the same sensitivity as the REB.

            This serves as an interesting example of a rapid 2- to 3-fold decline in low-magnitude seismicity within a very large geographical region, especially when compared to the subsequent 38-hour REB-event-free period between the morning of 2025197 and the late evening of 2025198. During this latter period, all curves for the weak LA version rise and then fall back to their initial levels. In this case, the strict/0.25 s curve also demonstrates growth, albeit emerging from a trough at the end of 2025196. 

Figure 28. The evolution of the running MS(4) of the quarter-day number of XSEL events between  2025187  and 2025201. The weak (upper panel) and strict (lower panel) with six origin time tolerance cases (from 5.0 s to 0.25 s) each are shown. Several intervals of with significant variations are marked by ovals.

 

            Figure 29 compares two LA versions and two cases, highlighting the periods of synchronous and asynchronous behaviors across various version/case permutations. When seismicity fluctuates simultaneously at all magnitude levels, the corresponding ratios evolve in sync.  Conversely, the asynchronous onset of growth and decline may be attributed to waves of low-magnitude seismicity passing through the magnitude thresholds that correspond to the respective version/case order numbers. The weak LA curves are nearly synchronous, with the exception of the final quarter-day (201.125, representing the first six hours of July 20). Specifically, the weak/0.25 s curve, which remains systematically below the weak/0.5 s curve, exhibited a sharp increase to align with the 0.5 s curve level. The strict curves did not mimic this behavior; consequently, the weak-to-strict ratio for the 0.25 s tolerance increased, as illustrated in Figure 30.  This is an example of low-seismicity progress through the magnitude range confined between the weak and strict versions.

           

Figure 29. Comparison of the 0.25 s and 0.5 s cases illustrating the synchronous in asynchronous behavior of the two LA versions and two cases.

 

Figure 30. The ratio of weak-to-strict LA versions for six origin tolerance cases between 2025187 and 2025203. The event at 21:38:01 on July 14 is marked by a black vertical dashed line.

 

Periods of significant decline in the strict/0.25 s curve compared to the weak curve were observed several times during the period between 2025187 and 2025201. The weak-to-strict ratio rose to high values, albeit against the background of decaying seismic activity. A prominent spike reaching a value of 38 occurred in the 0.25 s ratio curve in the second quarter-day of 2025189, when the XSEL count for the strict version fell to 1. A broader peak was observed on 2025195, when the strict curve dropped to a level of 3. Similarly, a sharp peak in the fourth quarter-day of 2025196 was induced by an abrupt decline in the strict LA version. This specific point marked the onset of growth in the strict version at a faster rate than in the weak version. This corresponding peak likely indicated the transition to the event on July 20.

            The two weeks preceding the J20 earthquake demonstrate a variety of peaks and troughs in the MS(4) of quarter-day XSELs and in the ratio of the weak and strict MS(4) XSELs. These peaks are most prominent in the curves for the 0.25 s origin-time tolerance, which defines the magnitude range closest to the corner magnitude of the REB recurrence curve and generates the most statistically significant XSEL event hypotheses. The XSEL ratio for this specific case may reveal the preparation for an impending earthquake, provided it is induced by an increasing number of events in the weak LA version, as was previously observed for the Sea of Okhotsk earthquake.

An ex post analysis allows for the differentiation of two distinct classes of peaks in the ratio curves. The first class is associated with a rapid drop in the number of events in the strict XSELs, accompanied by a much slower decline or no decrease at all in the weak XSEL version. The resulting peaks are sharp and exhibit large amplitudes, which are likely associated with periods of extremely low seismic activity. Statistically, recording only one to three events per day within a highly seismic region of a size of approximately 2000 km × ~1400 km is inherently characterized by statistical uncertainty. In contrast, the peaks in the weak-to-strict ratio curves observed prior to the J20 earthquake are supported by more robust statistics within the strict LA version.

The 0.5 s ratio curve also demonstrates peaks in nearly the same time intervals; however, these peaks are systematically lower, whereas the underlying MS(4) XSEL curves remain higher than the 0.25 s curves. The 0.5 s ratio curve exhibits a peak along the growth trend two quarter-days prior to the J20 earthquake and then decreases slightly. This behavior differs from that observed before the Sea of Okhotsk earthquake, where the 0.5 s curve began to rise a few quarter-days before the 0.25 s curve, after which both curves peaked synchronously right before the mainshock. This sequence was interpreted as a wave of events with an increasing magnitude progressing through a set of magnitude thresholds defined by the increasing order number of the LA version/origin-time tolerance case.

The four other curves also exhibit minor shifts toward the peaks of the 0.25 s curve. There is an episode of countermotion near 2025189, where the other five curves form a trough while the 0.25 s curve approaches its peak value. The significance of XSEL bulletins with the largest origin time tolerances depends on the sensitivity of the observational networks and the magnitude of the earthquake under study. For the Sea of Okhotsk earthquake, all six XSEL ratio curves contributed specifically to estimating the level and tracking the progression of the seismic activity wave from the lowest magnitudes to the mainshock. In contrast, for the study of the Kamchatka earthquakes, they likely serve only as a baseline reference, assisting in the reconstruction of the recurrence curve, such as those presented in Figure 18. 

 

Seismic activity between the J20 and J29 earthquakes

The period between the J20 and J29 earthquakes is of special interest because it involves seismic activity that is significantly higher than that observed prior to the July 20 earthquake. Although the J20 aftershock sequence decays rapidly toward the J29, as illustrated in Figure 19, it could pose a potential challenge for the WCC processing to distinguish between the REB matching and new XSEL events before the J29 mainshock. While the REB matching statistics in Table 4 illustrate the efficacy of the WCC in detecting REB events, it remains a question whether the WCC can differentiate between the mixture of low-magnitude J20 aftershocks and J29 foreshocks, or if this mixture represents a single, continuous process of earthquake preparation. Up to this point, the XSELs included all events, encompassing both the REB-matching and newly detected ones. The latter subset of XSEL events does not differ significantly from the total version for quiet days, and could therefore be utilized instead of the total number for earthquake prediction purposes.

The period between the J20 and J29 events, however, is not characterized by seismic quietness. Figure 31 compares the total number of events with the number of new XSEL events for both the weak and strict LA versions under the 0.25 s and 5.0 s origin-time tolerance cases. The difference between these curves represents the number of REB-matching XSEL events. It rises across both versions and for both cases. Within the strict version, the gap between the 0.25 s and 5.0 s cases is small on the first day after the J20 event and practically disappears after the J29 mainshock. For the weak version, the 5.0 s total and new curves virtually coincide after 2025207, whereas the 0.25 s curves converge only two days prior to J29. In contrast, the 0.25 s strict total and new curves retain the gap until the final minute prior to the J29. Overall, because the total XSEL counts inherently include J20 aftershocks up to the onset of the J29, these aftershocks and the J29 foreshocks cannot be distinguished. Nevertheless, the J20 earthquake failed to follow the J29 evolution steps, suggesting that the overall preparation for the mega-earthquake may not be disrupted by a relatively small loss of the total elastic energy accumulated in the entire region in J20 mainshock and its aftershocks.

 

Figure 31. Evolution of weak (upper panel) and strict (lower panel) MS(4) numbers in the total XSEL vs. only the new events XSEL. Two origin time tolerances are used: 0.25 s and 5.0 s.

 

The evolution of the MS(4) for the  total XSEL curves between 2025200 and 2025211 is presented in Figure 32. Six curves representing all cases within the same versions move nearly synchronously, displaying only a few minor deviations. Furthermore, curve pairs sharing the same origin-time tolerance case appear highly similar across both LA versions. The principal divergence among the curves in Figure 32 is observed during the day preceding the J29 mainshock. The 0.25 s curve for the weak version demonstrates a clear upward trend, whereas the strict version exhibits a downward trend common to all six cases, punctuated by a minor  peak two quarter-days prior to the J29 rupture.

The weak-to-strict ratios for the six origin time tolerance cases in Figure 33 illustrate a synchronous evolution of the MS(4) metric, characterized by an overall positive trend from 2025200 to 2025211. The most pronounced variations in these ratios are confined to the 0.25 s and 0.5 s cases. This pattern is common to all events analyzed thus far: the Sea of Okhotsk, the J20, and the J29 events, as well as the two quiet periods. Following the J29 earthquake, all six ratio curves drop sharply to a baseline level of 2, a feature that is directly linked to the abrupt drop in the MS(4) levels illustrated in Figure 20. Because of the smoothing effect inherent to the running sum window, the first three quarter-days following the mainshock still incorporate MS(4) values from the pre-seismic interval, thereby smoothing the transition to the lower post-seismic baseline. A ratio value of 2 corresponds to a regime where the number of newly detected XSEL events is negligibly small compared to those matching the REB bulletin. In other words, the magnitude threshold for the WCC processing is shifted very close to the corner magnitude of the total recurrence curves in Figure 15 or 17, depending on the respective depth estimates in the REB and XSEL.

 

Figure 32. The MS(4) curves between 2025200 and 2025211.

 

The weak-to-strict ratios for the 0.25 s and 0.5 s tolerance cases in Figure 33 rise from a level of 4.0-4.5 up to 7.0 during the last two to three quarter-days prior to the J29. This peak is observed within the last 5 hours preceding the mainshock, which occurred at 23:24:48 (IDC origin time). The last hour of 2025210 was excluded from the pre-event WCC processing. This last hour was also omitted from the post-seismic processing because both the REB and XSEL are severely biased by extraordinary noise levels. There is a study devoted to the recovery of aftershocks in the initial minutes following the J29 [Kitov et al., 2026]. The exclusion of this hour should not introduce any measurable bias into the weak-to-strict version ratio, as it affects both versions proportionally and exhibits no discernible influence on the three cases ranging from 2.0 s to 5.0 s. The 1.0 s case demonstrates an increase, peaking near 5.5.

 

Figure 33. The ratios of the number of XSEL events in the weak and strict LA versions for the period of time between 2025200 and 2025215.

 

The peaks in the 0.25 s and 0.5 s curves immediately preceding the J29 mainshock, with the depth fixed at the free surface, represent a potential short-term earthquake prediction trigger. A similar precursor was previously observed prior to the Sea of Okhotsk M8.3 earthquake.  The consistent behavior of low-magnitude seismicity in both cases provides compelling evidence in favor of such a hypothesis.

             

Recurrence curves for the ME sets

            The best 100 MEs constituted the initial configuration covering the entire studied region. Thus far, this specific set has been utilized in the calculation of the XSEL bulletins, as well as the corresponding time series of the MS(4) metrics and the weak-to-strict ratios. There is another ME set consisting of 195 REB events, each featuring a mandatory associated P-wave arrival at station PDYAR. This latter set also covers the same geographical area and depth range but was selected from the REB after 2023, following the upgrade of this station to an operational array at the end of 2022. A cumulative total of 295 MEs were utilized to create the XSEL events; both constituent sets are illustrated in the left panel of Figure 13.

The expanded set allows for the investigation of several critical problems related to the precursory peaks observed in the 0.25 s and 0.5 s weak-to-strict ratio curves in Figures 30 and  33. First, it is necessary to determine which of the two configurations exhibits greater efficiency in generating XSEL events, given their subtle variations in spatial distribution and sample size. The top 100 MEs are likely characterized by superior sensitivity and resolution, a feature supported by their selection procedure, which relies on a comprehensive pairwise WCC analysis with neighboring REB events. The ME set anchored by station PDYAR is larger and directly benefits from the high sensitivity of this array to seismic events within Flinn–Engdahl region 19 (Kuril–Kamchatka).

The subsequent question involves resolving the geographical provenance of the XSEL events that drive the precursory trigger shown in Figure 33. The earthquake preparation processes may structurally encompasse the entire regional volume bounded by coordinates 45°N – 65°N, 145°N – 165°N and a depth range from 0 km to 700 km. Alternatively, the preparation zone can be restricted to a small spot, such as the J20 aftershock sequence in Figure 12, concentrated in the lithosphere, or a broader spatial envelope corresponding to the J29 aftershock distribution. To test these hypotheses, the total 295 ME set is partitioned into four distinct geographic domains in the right panel of Figure 13:

·       The entire region including the subducting Pacific plate.

·       The localized zone surrounding the potential asperity that arrested the slip propagation of the J20 event along the eventual J29 rupture line (hereafter referred to as the “Asperity” zone, which excludes a single deep-focus event within its boundary and incorporates 49 MEs).

·       The remaining spatial domain outside this core zone, designated as “Out of Asperity”.

·       The southwestern periphery situated beyond the immediate J29 aftershock zone, designated as the “Rim” region, which utilizes 79 MEs from the “PDYAR” configuration.

 

This spatial subdivision is baseline-exploratory, aimed at identifying potential regional accents in the precursory signal.  If necessary, this tentative division can be optimized during a subsequent technical calibration procedure. One such optimization parameter involves the recurrence curve, which characterizes the performance of the WCC pipeline in terms of how accurately the order numbers of specific version-case pairs correspond to a uniform sequence of magnitude thresholds. The upper panel of Figure 34 illustrates this parameter computed for the period between 2025193 and 2025201. The curve for the best 100 MEs is identical to that presented in Figure 16, whereas the «JOINT» curve represents the complete set of 295 MEs. The «JOINT» bulletin is not a simple cumulative sum of the best 100 and 195 “PDYAR” XSELs; rather, it is the direct product of joint processing incorporating a Conflict Resolution (CR) algorithm those cross-rejects redundant XSEL events between the two constituent bulletins. For instance, the «JOINT» XSEL yields 3752 events in the v₁c₁ pair (weak version, 5.0 s tolerance), whereas the standalone best 100 MEs generate 2625 events and the  “PDYAR” configuration produces 2642 events. For the v₂c₆ pair (strict, 0.25 s), the event counts are 140, 74, and 92 events, respectively. Thus, the value added to the «JOINT» XSEL bulletin scales from 50% for the v₁c₁ pair to nearly 100% for the v₂c₆ pair. 

The respective curves in the upper panel of Figure 34 evolve synchronously, displaying similar low-amplitude deviations within the weak LA version. However, the strict LA version for the “PDYAR” subset generates larger numbers of XSEL events across all six tolerance cases. This behavior is likely attributed to the larger number of MEs physically encompassing a broader spatial domain during the high-resolution grid search, which operates with a maximum radius of 43.2 km in the strict LA version.  For the weak version, the grid radius is 81 km, and the neighboring MEs have overlapping search areas leading to the CR process. This phenomenon is important for the future ME setting – the number of MEs should be sufficient to cover the whole region without detection-creation blind zones. The “PDYAR” master subset was partially introduced to address this issue, and the «JOINT» set is likely an important step on the way to an optimal ME number and distribution. The increasing number of MEs can yield additional XSEL events, but it would require higher computing power or longer processing time.

 

            There are minor deviations from the «JOINT» XSEL exponential regression line within the strict version XSELs. The lower panel of Figure 34 presents the recurrence curves from the upper panel, normalized to their respective maximum values for the v₁c₁ pairs. These relative variations effectively highlight the outliers across the configurations. The highest coefficient of determination, R²=0.9805, is yielded by the top 100 MEs set. This value is statistically significant enough to validate the correlation between the order numbers and the uniform sequence of magnitude thresholds. This robust correspondence represents a key feature underlying the reliability of the weak-to-strict version ratio as a predictive parameter.  The largest fluctuations are demonstrated by the “Rim” curve with R²=0.94. This specific curve exhibits a progressive deficit in XSEL events when approaching the highest order numbers. Both regression lines have the same slope of -0.31.

 

Figure 34. Recurrence curve characteristics evaluated across distinct Master Event (ME) sets. Upper panel: The recurrence curves for the 12 LA version/tolerance case pairs computed for different ME configurations. Lower panel: The corresponding recurrence curves from the upper panel, normalized to their respective absolute levels in the v1c1 configuration to highlight relative morphological variations and outliers.

 

Evolution of the number of XSEL events

            Because the weak-to-strict version ratio represents a measurable parameter designed to serve as an operational trigger for a major earthquake alert, the fundamental properties of the underlying calculations become crucial. For each monitoring interval, six distinct ratio curves are evaluated, corresponding to the six origin-time tolerance figures ranging from 0.25 s to 5.0 s. These metrics are derived from the respective XSEL bulletins generated by the WCC detection and LA procedures for individual MEs. The aggregation of these individual XSELs into a consolidated final XSEL is governed by a CR procedure, which is finely calibrated against the baseline daily event rates observed during the periods between major earthquakes. The backbone of this entire WCC processing pipeline comprises a spatial network of MEs combined with waveform templates of the first P-wave recorded across all stations associated with each respective ME.

            A standalone ME creates event hypotheses strictly within the spatial grid search radius relative to its own hypocentral position reported in the REB. While the probability of detecting an event situated outside this grid boundary is non-zero, its reconstructed location will be artificially shifted inside the grid.  This artifact can introduce significant errors in the absolute location, a limitation that is highly common in automatic bulletins. The REB events are also inherently mislocated relative to their true positions; however, the confidence ellipses derived from travel time residuals provide an estimate of the 95% probability contour containing the actual source. In general, a larger number of associated stations reduce the dimensions of this confidence ellipse. These structural parameters, alongside potential phase misassociation, render the choice of MEs highly vulnerable to location errors and template quality.

            The set of 295 MEs includes REB events of varying quality regarding absolute location accuracy, the number of associated stations, and the availability of valid templates of the first P-phase. The operational efficacy of individual MEs can vary significantly across the studied region. This variability can introduce measurable fluctuations in both the resolution and sensitivity of the WCC processing, directly affecting the final XSEL bulletins. Smaller ME subsets assigned to specific geographical areas may exhibit localized underperformance due to poor ME quality, which could be misinterpreted as low seismic activity. Genuine spatial fluctuations in seismic activity across the selected could be erroneously interpreted as instrumental or algorithmic underperformance.

            The partitioning into these geographical subsets is accomplished using simple straight lines that delineate quasi-rectangular areas. The probability space within which an individual ME can generate a valid event hypothesis can readily extend into adjacent zones. A discrete XSEL bulletin can incorporate events that geographically belong to a different ME subset. Within the complete ME network, however, these overlapping MEs would generate competing hypotheses that are ultimately resolved by the CR procedure. For example, the “Out of Asperity” domain contains numerous MEs situated immediately along the boundary of the “Asperity” core zone. This spatial overlapping introduces a non-zero bias in the independent XSEL estimates computed for each discrete subset. Within the integrated «JOINT» master set, this specific spatial bias is successfully suppressed by the presence of the MEs that explicitly belong to the “Asperity” set.

            The operational performance of all individual ME subsets can be systematically compared to the «JOINT» set as an aggregated output of all MEs in all subsets. Panel a) in Figure 35 depicts the six weak-to-strict ratios between 2025193 and 2025201 for the «JOINT» set. There are notable differences observed in the behavior of the 0.25 s and 0.5 s curves when compared to the best 100 MEs subset in Figure 30. The precursory peak amplitude in the «JOINT» 0.25 s curve decreases to 13.1, compared to 15.4. The double-peak structure preceding the J20 event in Figure 30 is smoothed, and the highest peak moved to the first place in their sequence. The peak at 2015199 lost its relative amplitude to 8.8 from 14.7 for the best MEs 100 set. The third peak near 2025197 is the highest for the presented interval. The period before 2025194 is not shown as the calculations for the «JOINT» set were conducted from 2025193 to 20252012. The 0.5 s curves are nearly the same for both sets with slightly lower amplitude relative to the highest peak. The remaining four tolerance curves are almost identical across both sets.

Overall, the 0.25 s curve has not lost its predictive capability, despite the fact that the weak MS(4) curve in Figure 36a exhibits only a single upward shift prior to the J20 event, while the strict MS(4) curve declines at a faster rate than the weak curve, followed by a sharp increase along the final leg preceding the J20 rupture, which is marked by the vertical red line. For the «JOINT» curve, the broader morphology of the precursory peak is driven primarily by an accelerated decline in the strict curve, mimicking the behavior observed during previous peak intervals. This underlying mechanism differs significantly from the forcing factors that govern the standalone best 100 MEs set.

The “PDYAR” subset in Figure 35b is characterized by a broader precursory peak that begins to grow three days before the J20 mainshock. The prominent trough near 2025200, which is clearly observed in the best 100 and «JOINT» ME sets, loses its depth in the “PDYAR” subset, appearing instead as a smooth transition between two individual maxima. Both of these peaks are driven by a faster decline in the strict curve relative to the weak LA version. The final quarter-day preceding the J20 event is characterized by a rapid amplitude rise in the strict curve – signaling an increasing activity of the XSEL events closer to the corner magnitude of the REB recurrence curve – which occurs alongside a more gradual increase in the weak curve. The peak level of the 0.25 s curve is 16.8, strongly supporting its precursory character.  The 0.5 s curve begins to decline three quarter-days prior to the J20 earthquake, consistent with the behaviors documented in the other two ME sets. This synchronous signature further reinforces the hypothesis of an impending major earthquake.

The 0.25 s ratio curve for the “Asperity” ME subset in Figure 35c exhibits features similar to those of the “PDYAR” subset, albeit with a few distinct deviations. The “Asperity” curve also starts to rise on 2025198, but reaches its peak value of 14.5 five quarter-days prior to the J20. Unlike in the «JOINT» and “PDYAR” subsets, there is a peak near 2025195 induced by a deep trough in the strict curve. There is a similar peak in the best 100 MEs curve. This empirical observation suggests that the sharp declines in the strict curves are driven by local conditions and depend on individual ME quality and spatial distribution, rather than encompassing the entire region. Such local peaks are effectively smoothed by the cumulative activity in the «JOINT» set. Only the largest anomalies remain visible in the curves that cover the entire region. For instance, the prominent peak near 2025197 is clear for the best 100, “PDYAR”, and «JOINT» sets, yet it remains statistically significant within the “Asperity” curve. Furthermore, the final precursory peak preceding the J20 event is significantly smoothed and displays a lower amplitude relative to the other maxima analyzed above.

The “Out of Asperity” 0.25 s ratio curve in Figure 35d demonstrates two distinct high-amplitude peaks: one near 2025197 and another three quarter-days prior to the J20. The latter maximum reaches an absolute value of 21.4. It could serve as an optimal short-term precursor to the J20 earthquake, especially given that its culmination is immediately followed by a systematic two-quarter-day decline leading directly to the J20 origin time. This peak is driven by a fast decline in the strict curve in Figure 36d, followed by an equally accelerated recovery that initiates three quarter-days before the J20. The growing number of events in the strict version XSEL is a feature potentially related to earthquake preparation. The geographical distribution of MEs within the “Out of Asperity” set suggests that the seismic activity prior to the coseismic phase was concentrated outside the zone of the “Asperity” subset. 

            The “Rim” subset is presented in Figure 35e. It was designed to cover the area beyond the aftershock sequence of the J29 mega-earthquake. It should inherently remain insensitive to any seismic activity related to the preparation of the J20 and J29 earthquakes. It has an extremely high peak near 2025199. While similar peaks are observed across the sets covering the entire region, they are much lower in relative amplitude, thereby confirming the strictly localized character of this anomaly. The “Rim” set in the southwestern periphery of the region undergoes its own independent seismic activity on 2025197, which should not be factored into the core J20/J29 nucleation process. There is another sharp peak with an amplitude of 42.0 four quarter-days before the J20. Its amplitude is directly related to the drop in the strict curve to the level of 1 to 2 per day, as illustrated by the corresponding MS(4) curves in Figure 36e. Despite its high amplitude, this peak is statistically insignificant due to the extremely small sample size. Mechanically, this anomalous spike can be attributed to the over-the-border spatial sensitivity of MEs positioned near the primary preparation zone of the J29 earthquake.

            a)

b)

c)

d)

e)

Source: http://mechonomic.blogspot.com/2026/06/prediction-of-kamchatka-july-29-2025_01434436280.html