Navigating the Storm: A Diagnostic Analysis of QZSS CLAS Performance and Vulnerabilities Through the Solar Cycle 25 Maximum

4.1.1 Spatially Dependent Accuracy

Our preliminary study (Shiono & Kubo, 2025) identified a spatial dependency in CLAS performance based on user location relative to grid points. This methodology defines the “inside network area” in two steps: first, individual grid points are classified as “inside” if they are within 30 km of the nearest CORS, and second, the “inside area” is defined as the geographical region covered by those points (as illustrated in Figure 4(a)). We confirmed this spatial dependency over the extended 2021–2025 Q3 analysis period. As shown in Figure 4(b), with the exception of a few outliers, the data points consistently lie on or above the y = x line of equality. This result confirms that the positioning error “outside” the network is almost always greater than or equal to the error “inside,” proving that proximity to the underlying CORS network remains a critical performance factor, irrespective of the solar cycle phase.

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 4

Analysis of the inside and outside network areas: (a) definition of “inside” (circles, ≤30 km from nearest GEONET station) and “outside” (crosses, >30 km) grid points, where polygons indicate inside-network areas; (b) 95th percentile of kinematic horizontal/ vertical accuracy and TTFF inside vs. outside network areas (2021–2025 Q3)

Points represent individual days, colored by season (summer: orange; winter: blue; other: gray); solid black line = y = x.

4.1.2 Comparison with Official Performance Report

This section compares our results with the semi-annual CLAS Service Performance Reports published by QSS, which use a fixed set of 72 core stations across Japan (Figure 1(b); temporary substitutes were employed for a few sites affected by the 2024 Noto Peninsula earthquake (QSS, 2025b)). The comparison covers January 2021–March 2025, the period for which official reports are available, and aims to validate our methodology while highlighting the influence of processing differences (e.g., 30-s vs. 1-Hz data, 4 days/month vs. daily evaluations).

Figure 5 shows close agreement between the two analyses in both static and kinematic modes from 2021 through mid-2022. From late 2022 onward, however, our results exhibit greater and more variable degradation than the official reports, with the largest divergence occurring during the 2024 solar maximum. Nevertheless, both data sets capture the same overall trend: pronounced worsening in 2024 Q3–Q4 followed by partial recovery in 2025.

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 5

Comparison of CLAS positioning accuracy from this study with the QSS Official Service Report (2021–2025 Q1) This figure compares the CLAS positioning accuracy results from this study with those reported in the QSS Official Service Report from 2021 to 2025 Q1. The performance standard thresholds are not shown, as a direct comparison against official criteria is considered inappropriate because of methodological differences (30-s vs. 1-Hz sampling). The left side presents static mode results, while the right side displays kinematic mode results. The top row shows the horizontal accuracy, and the bottom row shows the vertical accuracy. The evaluation stations are limited to the 72 core stations used in the official report, as shown in Figure 1(b).

These discrepancies arise primarily from methodological differences. The official 1-Hz, daily evaluations on a small, fixed station set yield smoother and generally better statistics, particularly under high solar activity, whereas 30-s sampling is more susceptible to short-term disturbances. Despite these limitations, our nationwide, 30-s analysis reproduces the major performance trends observed in the official reports, confirming the effectiveness of our analysis for identifying long-term and system-level behaviors that may be averaged out in the official metrics.

4.1.3 Breakdown of Accuracy Degradation Factors

This section provides a breakdown of the factors contributing to the accuracy degradation observed in CLAS performance from 2021 to 2025 Q3. Figure 6 illustrates the decision process used to classify the primary cause of each degradation event. The classification employs quantitative thresholds (e.g., Dst < –150 nT for geomagnetic storms, CS count > 100 for PB-related events) combined with spatiotemporal alignment verification using detrended TEC (dTEC)/rate of TEC index (ROTI) maps, CSSR corrections, and GNSS observation quality metrics (Nose et al., 2015).

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 6

Flowchart for classifying primary causes of accuracy degradation events

This flowchart illustrates the simplified decision process used to identify the primary cause of each degradation event (daily 95th percentile error exceeding the performance standard). The CS threshold (100) refers to the sum of mean daily CSs per station across all compact networks, as shown in the upper panel of Figure 9(a). The classification was performed independently for each affected compact network; consequently, a single day may be associated with multiple contributing factors across different networks (e.g., ionospheric and tropospheric). Although the flowchart depicts a sequential decision tree, in practice, all available data sources–including dTEC/ROTI maps, CSSR corrections, and GNSS observation quality metrics–were comprehensively examined for each event, regardless of the branch taken. Cases in which multiple indicators (e.g., wave-like patterns and elevated CSs) were present simultaneously were resolved through spatiotemporal alignment analysis to determine the dominant mechanism.

Figure 7(a) summarizes the annual rate of accuracy degradation events and their severity levels over the analysis period. An “accuracy degradation event” is defined as a day when the 95th percentile horizontal or vertical error exceeded the performance standards in either static or kinematic mode. The severity levels were classified into two categories: “Severe” and “Moderate.” If either mode exceeded 125% of the performance standard, the event was classified as “Severe”; otherwise, it was classified as “Moderate.” During the quiet- or moderate-solar activity years of 2021 and 2022, the degradation event rate was relatively low, at 4% and 13%, respectively. However, as solar activity intensified in 2023 and peaked in late 2024, the event rate surged to 53% in 2023 and reached a maximum of 62% in 2024. In 2025 Q1 to Q3, as solar activity began to wane, the event rate decreased to 51%. The ratio of “Severe” and “Moderate” events remained relatively consistent throughout the years, with “Severe” events accounting for approximately 44.9% of the total degradation events.

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 7

Breakdown of accuracy degradation events from 2021 to 2025 Q3: (a) annual count of accuracy-degradation events (95th percentile exceeding standards in static or kinematic mode), where blue = severe (>125% of standard), red = moderate; (b) primary causes of degradation events (ionospheric, tropospheric, unidentified); (c) severity composition for each ionospheric disturbance type

Figure 7(b) shows the primary cause of degradation events. To identify the root causes, we conducted a comprehensive analysis using various data sources, mainly including positioning error time series, GNSS observation data (SNR, MP, CS), GEONET information (GSI, 1996), ionospheric disturbance maps such as the dTEC and ROTI (National Institute of Information and Communications Technology [NICT], 2025; Saito et al., 2014, 2021), and space weather information (NOAA SWPC, 2025; World Data Center for Geomagnetism, Kyoto et al., 2015). A single event can have multiple contributing factors; however, this analysis focuses on classifying each event by its primary cause. The degradation events were primarily categorized by their likely source: ionospheric, tropospheric, or unidentified disturbances. The ionosphere-related events were further classified into five specific categories based on their characteristics: medium-scale traveling ionospheric disturbances (MSTIDs), PBs, geomagnetic storms, and other complex disturbances. The analysis revealed that the dominant cause of accuracy degradation was ionospheric effects for all years considered, accounting for 93.3% (83 out of 89) of the total events. The most dominant event type was PBs (28 events, 31.5%), followed by MSTIDs (27 events, 30.3%), other/complex disturbances (25 events, 28.1%), and geomagnetic storms (3 events, 3.4%). The findings are summarized as follows:

• MSTIDs: MSTIDs are a well-known phenomenon involving the propagation of wave-like fluctuations with periods ranging from 15 to 60 min and horizontal wavelengths of several hundred kilometers (Hocke & Schlegel, 1996; Hunsucker, 1982). In Japan, MSTIDs are frequently observed during summer night and winter daytime, with a clear seasonal and solar cycle dependence. The occurrence rate of both daytime and nighttime MSTIDs is reported to decrease with high solar activity (Otsuka et al., 2021).

  Note that the MSTID events listed here are those that caused significant degradation in CLAS performance, not all observed MSTID events. From 2021 to 2022, during low to moderate solar activity, MSTID-induced degradation events consisted of 4 nighttime and 3 daytime events. However, as solar activity intensified in 2023 and peaked in 2024, the number increased to 7 nighttime and 11 daytime events. This apparent contradiction—fewer MSTIDs overall yet more degradation events—can be explained by distinguishing between relative and absolute disturbance magnitude. While Otsuka et al. (2021) define MSTID activity as a relative metric (δI/I¯, perturbation divided by background TEC), the absolute TEC fluctuation (δI) increases substantially during a solar maximum owing to the elevated background TEC. Because GNSS positioning errors are driven by absolute phase shifts rather than relative ratios, the MSTID events that do occur during high solar activity are more likely to exceed the system’s robustness threshold, causing significant degradation. Furthermore, these severe MSTID events triggered system instabilities, leading to both degradation and instability in the positioning accuracy. This phenomenon will be discussed in more detail in Section 4.3.3.

• PBs: Equatorial PBs are known to be more frequent after sunset, typically from Coordinated Universal Time (UTC) 10:00 to 15:00 (Japan Standard Time 19:00 to 00:00), particularly during equinoxes and periods of high solar activity (Groves et al., 1997). These PBs often reach mid-latitude regions, including Japan, especially during solar maximum periods. This leads to significant scintillation and phase fluctuations, often manifesting as CSs in GNSS signals, severely impacting positioning accuracy. Our analysis indicates that PB-induced degradation is a major factor, accounting for 31.5% (28 out of 89) of all degradation events. Furthermore, the data clearly support the correlation with solar activity, showing an increasing trend both in the number of PB-related degradation events and the number of CSs observed at GEONET stations as solar activity intensified.

• Geomagnetic Storms: Geomagnetic storms are primarily caused by Earth-affecting solar transients, specifically coronal mass ejections and stream interaction regions, with the key requirement for their generation being the presence of a strong and persistent southward magnetic field within these structures. These solar events and the resulting storms subsequently cause significant impacts in the near-Earth space environment, including ionospheric disturbances and ionospheric scintillation, which ultimately degrade GNSS signal performance. These phenomena are also closely tied to the solar cycle, with increased frequency and intensity during solar maximum periods. Although infrequent, geomagnetic storms were responsible for one major degradation event in 2023 and another in 2024.

• Other/Complex Disturbances: This category encompasses ionospheric disturbances that do not neatly fit into the other classifications. These disturbances may include events with mixed characteristics or those that are not well understood. The number of these events increased from 2 in 2022 to 8 in 2023 and 9 in 2024, indicating that the complexity of ionospheric disturbances increased as solar activity intensified.

Figure 7(c) illustrates the ionosphere-related degradation composition by severity level. The severity level distribution includes more uncorrectable events caused by CSs, such as PBs, compared with MSTIDs and geomagnetic storms. This trend suggests that certain uncorrectable ionospheric disturbances have a more pronounced impact on CLAS performance.

Troposphere-related degradation events are the second-most prevalent category, accounting for 5.6% (5 out of 89) of the events. If we include cases in which tropospheric issues occurred concurrently with other factors, the total number of related events rises to 11.2% (10 out of 89). A notable finding is that these events were not aligned with physical tropospheric conditions (e.g., heavy rainfall), suggesting that these events are symptoms of system-level vulnerabilities. This hypothesis is supported by two key findings from our preliminary analysis (Shiono & Kubo, 2025). First, the event on October 1, 2021, showed that a large, unmodeled SIS error (a QZSS J03 orbital bias) was incorrectly absorbed by the tropospheric estimation model, causing it to diverge. Second, CLAS-derived tropospheric estimates exhibit systematic, network-dependent biases when compared with the GSI F5 final solution, suggesting that these biases stem from the integration of such contaminated local estimates. This issue will be discussed in more detail in Section 5.2. Notably, for the later years of the analysis period, we observe a decreasing number of satellites used in positioning concurrent with these tropospheric events, which will be discussed in Section 4.3.1.

Finally, there was only one unidentified disturbance event (1.1%) during the entire analysis period, occurring in 2023. This event did not exhibit any clear ionospheric or tropospheric signatures. A possible explanation involves degraded GNSS measurements from low-elevation satellites; when these satellites were excluded from the solution, the positioning accuracy returned to normal levels. However, the root cause remains difficult to determine from the available data.

4.2.1 Time of Day and Season Analysis

This section delves into the influence of ionospheric disturbances on CLAS performance, particularly focusing on how these effects vary with the time of day and season. Major ionospheric disturbances such as MSTIDs and PBs exhibit distinct temporal and seasonal patterns, which can significantly impact the GNSS positioning accuracy and TTFF. To investigate these effects, we analyzed the positioning accuracy and TTFF data from 2021 to 2025 Q3, categorizing the results based on daytime and nighttime conditions as well as seasonal variations. The daytime and nighttime classifications are based on UTC time, with daytime defined as 00:00–09:59 and 21:00–23:59 UTC and nighttime defined as 10:00–20:59 UTC.

Figure 8(a) illustrates the relationship between daytime and nighttime positioning accuracy across different seasons. The characteristic and seasonal patterns of PBs are evident in the gray points, as they are widely scattered around the y = x line. The orange points (summer season) are predominantly above the y = x line for both horizontal and vertical components, clearly indicating MSTID characteristics. In contrast, the blue points (winter season) cluster into two groups: around the y = x line or widely scattered, with the latter coinciding with the solar maximum. During periods of low solar activity (2021–2022), the blue points cluster closely around the y = x line; however, as solar activity intensifies (2023–2024), the points become more widely scattered, indicating increased variability in accuracy between daytime and nighttime. This trend is also reflected in the TTFF performance shown in Figure 8(b).

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 8

Impact of ionospheric delay, based on time of day and season analysis in kinematic positioning (2021-2025 Q3): (a) daytime vs. nighttime 95th percentile kinematic accuracy (points = individual days, colored by season; solid black line = y = x); (b) complementary cumulative distribution function (CCDF) of kinematic TTFF for summer (left) and winter (right), comparing 2021 (blue) and 2024 (orange); solid = daytime, dashed = nighttime

Figure 8(b) reveals the temporal trend in TTFF performance, comparing 2021 (low solar activity) and 2024 (high solar activity). In the summer, the 2021 data (blue lines) show a clear performance gap: the daytime TTFF (solid line) is good, whereas the nighttime TTFF (dashed line) is significantly worse. As solar activity increases towards the 2024 maximum (orange lines), both daytime and nighttime TTFF performances degrade substantially, with the 95th percentile often hitting the 30-epoch limit. In the winter, the 2021 data show that the daytime and nighttime TTFF performances are nearly identical. However, by 2024, both daytime and nighttime TTFF performances degrade significantly, indicating that the overall ionospheric disturbance level had increased because of heightened solar activity, impacting both conditions. In particular, the degradation in the winter of 2024 appears to be more severe than that observed in the summer of 2024. This result suggests that these trends align with the timing of the solar maximum, which intensified the background TEC. This higher background TEC likely amplifies the absolute magnitude of fluctuations caused by daytime MSTIDs, leading to a more severe impact during winter months (as hypothesized in Section 4.1.3).

4.2.2 Impact of Increasing PB Frequency on CSs and Positioning Accuracy

The preceding section presented the overall trend of ionospheric disturbances, identifying MSTIDs as a dominant degradation factor. This section focuses on the specific impact of PBs on CSs and positioning accuracy.

Figure 9(a) presents the time series of mean daily CSs per station from 2021 to 2025 Q3. The number of CSs is notably elevated during spring and autumn each year, with a pronounced increase observed through 2023 and 2024, coinciding with the approach to the current solar maximum. This increase affected station observation performance, as shown in Table 1. The geographical distribution of CS activity expanded northward over the years, reaching the northern regions of Japan during the solar maximum (2024 Q3-2025 Q1). Notable increases in CSs occurred on November 5, 2023, and March 18, 2024. The most extensive northward expansion was observed on January 1, 2025, when significant CSs were recorded, even in the northern Tohoku region. Two of these events (November 5, 2023, and January 1, 2025) correspond to major geomagnetic storms, whereas the event on March 18, 2024, is attributed to PBs. The former event (November 5, 2023) is well known for producing red auroras at middle latitudes worldwide (Kataoka & Bamba, 2023). This event caused a marked increase in CSs (Figure 9(a)) and substantial degradation in positioning accuracy (Shiono & Kubo, 2025). The latter event (January 1, 2025) was also a major storm (Dst index below -200 nT). Other researchers have independently reported the expansion of PBs into northern Tohoku during that storm (Sakanoi, 2025), corroborating the trend observed in our data and linking severe high-latitude ionospheric disturbances to the solar maximum. Note that our data set is intermittent (one day per week); thus, other significant PB events may not have been captured in this analysis (e.g., May 11, 2024) (NOAA SWPC, 2024).

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 9

Impact of PBs on CSs and positioning accuracy: (a) time series (upper) and heatmap by compact network (lower) of mean CSs per station (2021–2025 Q3), with hatched backgrounds indicating winter (blue) and summer (green); (b) log-log relationship between mean daily CSs and 95th percentile kinematic 3D error in 2024 (point size ∝ 95th percentile dTEC); solid line = power-law fit; TECU: TEC unit In panel (b), Compact Network ID 12 data for 18 March 2024 were excluded owing to missing dTEC data.

Figure 9(b) illustrates the relationship between mean daily CSs, 3D positioning accuracy (95th percentile), and ionospheric disturbance for 2024. The size of each data point (bubble) represents the magnitude of ionospheric disturbance, quantified by the 95th percentile of daily dTEC for each network. To quantify the relationship between CSs and positioning error, we performed linear regression on the logarithmically transformed data. The coefficient of determination (R² ≈ 0.49) indicates that approximately 49% of the variance in the logarithm of positioning error can be explained by the logarithm of CSs. This correlation suggests that CSs were a major factor in the degradation of positioning accuracy during this solar-maximum period.

The power-law relationship between the two variables has a slope of 0.42 (y ∝ x^0.42), indicating that a 10-fold increase in CSs corresponds to a 10^0.42 ≈ 2.6-fold increase in positioning error. The regression line intersects the 3D performance standard of 26.8 cm at approximately 14.7 CSs, providing a practical threshold for performance degradation. These results highlight the challenges that solar-maximum conditions pose for high-accuracy GNSS positioning.

4.3.1 Decreasing Number of Augmented Satellites: Anomalous Drops And Long-Term Trends

The number of augmented satellites is an important factor in precise GNSS positioning performance. Although CLAS specifications (IS-QZSS-L6-003 through 006) support up to 17 satellites (CAO, 2020, 2025b), the daily mean number used in positioning solutions fluctuated between 12 and 15 throughout the study period, with a downward trend (Figure 2, top middle panel).

We identified two distinct phenomena:

• Long-Term Trends: Sustained decreases over several months in late 2021, late 2023, and late 2025 (Section 4.1).

• Anomalous Drops: Sharp reductions, occurring occasionally to as few as six satellites.

Figure 10 compares geometrically visible satellites (above 15° elevation), CSSR-augmented satellites, and satellites actually used in positioning. The key findings are summarized below.

1. Satellites used in positioning (green) closely track CSSR-augmented satellites (orange), confirming that rovers effectively utilize available augmentation. However, the number of augmented satellites remains consistently 2–5 below the number of geometrically visible satellites (blue), with ratios of 70%–90%.

2. The number of geometrically visible satellites increased steadily, yet the number of augmented satellites did not increase correspondingly, widening the gap over time.

3. The three sustained drops were primarily driven by reduced QZSS augmentation, as confirmed by the GNSS missing rate (third row, spikes in the QZSS green line). The bottom heatmap reveals the specific root causes. Throughout the study period, the active constellation consisted of four satellites; however, the geostationary Earth orbit (GEO) satellite QZS-3 (J07) was excluded owing to low dynamics. Consequently, the service effectively relied on only three usable quasi-zenith orbit (QZO) satellites (QZS-1/1R, QZS-2, QZS-4). Consequently, the unavailability of any single QZO satellite directly reduces the augmented count and constrains the overlapping visibility required for double-difference processing. The specific causes of each drop are as follows:

   • ⚬ Late 2021: The decommissioning of QZS-1 (J01) (NAQU2022059) and commissioning of QZS-1R (J04) (NAQU2021199, 2022011, and 2022052) (QSS, 2025d) created a transition period during which the visible QZO constellation was effectively reduced from three to two satellites (J02 and J03). Although the heatmap shows that J02 and J03 remained fully available, the total count of augmented satellites still decreased owing to the loss of the third QZO satellite. This finding suggests that the generation system maintained processing continuity by internally utilizing transitional satellites (J01/J04) or the GEO satellite (J07) as reference links, despite their exclusion from user-facing augmentation.

   • ⚬ Late 2023: No major QZSS outages were reported, yet drops occurred despite three QZO satellites being nominally available, indicating internal CLAS vulnerabilities.

   • ⚬ Late 2025: The QZS-1R L1C/A-to-L1C/B signal switch (NAQU2025351) caused issues because many GEONET stations lacked proper L1C/B support. Although QZS-6 (J08) had been commissioned (NAQU2025030 and 2025248) (QSS, 2025d), this satellite could not fully compensate for QZS-1R issues because it is a GEO satellite transmitting L1C/B. This situation effectively reduced the number of usable QZO satellites to two or fewer. Intermittent Galileo absences also contributed to these issues.

4. The 2025 drop (ratio falling from 78% to 67%) coincided with broader performance degradation (Section 4.1) and culminated in severe anomalous drops to 6–7 satellites on September 17 and 24, 2025, exacerbated by concurrent tropospheric divergence. The underlying long-term gap likely amplified these short-term events. The impact of such drops is officially acknowledged; for example, NAQU2023094 (QSS, 2025d) explicitly stated that CLAS performance in ID 01 (Ishigaki) was degraded because the satellite count fell below five.

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 10

Long-term trend of augmented satellites: (first row) mean counts of geometrically visible, CSSR-augmented, and used satellites across all networks; (second row) ratios to geometrically visible satellites by compact network; (third row) GNSS satellite missing rates for each GNSS constellation; (bottom row) heatmap of mean missing rate per QZSS satellite across all networks

4.3.2 Reference-Frame Mismatch: System Offset in Static Vertical Component

As noted in Section 4.1, the static-mode vertical accuracy remained elevated into 2025 without recovery, unlike the horizontal components and kinematic-mode results. We primarily attribute this result to reference-frame inconsistency between CLAS (ITRF2014-based GSI F5 coordinates) and IGS antenna PCV products, which switched from igs14.atx (ITRF2014) to igs20.atx (ITRF2020) on November 27, 2022 (IGS, 2021).

Daily east/north/up offsets were computed for each station as the difference between the static-mode daily median position and smoothed reference coordinates. Horizontal components remained stable near zero throughout the period. In contrast, the vertical (up) component showed a near-zero median in 2021-2022 but exhibited a widening spread and a positive median shift from 2023 onward, coinciding exactly with the IGS frame change. A slight positive shift also appeared in the east component from 2024. Mann–Whitney U-tests confirmed the statistical significance of these shifts (e.g., p ≈ 10⁻¹⁵ between 2021 and 2025 distributions).

Figure 11(a) reveals an increasing nationwide vertical offset trend, with regionally varying magnitude. When a 3-cm threshold (one-quarter of the 12-cm static vertical standard) is used, the number of affected stations increases from 156 (2021) to 466 (2025; Figure 11(b)). Affected stations showed no common receiver/antenna characteristics, confirming a systemic rather than hardware-specific issue.

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 11

Analysis of vertical offset in static mode (2021–2025): (a) spatial distribution of vertical offset slope; (b) annual count of stations with median vertical offset >3 cm; (c) box plots comparing backbone cluster (approximately 20 stations) and non-backbone stations

To further diagnose the origin of this bias, stations were divided into approximately 20 “backbone cluster” stations (core anchors of the GSI F5 solution (Muramatsu et al., 2021)) and the remaining non-backbone stations. Figure 11(c) shows that non-backbone stations developed offsets from 2023, whereas backbone stations remained stable through 2024. Backbone stations exhibited significant bias only in 2025, indicating that the reference-frame inconsistency progressively accumulated until it affected even the anchor network.

4.3.3 Case Study: Impact of Message Generation Facility Switching During Ionospheric Disturbances

This section examines the impact of message generation facility (MGF) switching on CLAS performance. CLAS operates two geographically independent MGFs, each with redundant server sets (four systems, numbered 0–3). Normally, one MGF serves as the primary facility, and its corrections are broadcast, while the others remain on standby (CAO, 2025b). Web archive data typically provide only the primary (system 0) corrections. However, during maintenance or failures, the active MGF is switched, and because each facility generates corrections independently, state continuity (e.g., ambiguity resolution) is not guaranteed. This situation forces receivers to reset states and re-initialize when inconsistent corrections are received.

A representative example occurred on February 5, 2025, amid widespread daytime MSTIDs across Japan (Figure 12(a)). Figure 12(b) shows the resulting kinematic positioning degradation at representative stations. Between 00:00 and 05:00 UTC, six MGF switches occurred; vertical dashed magenta lines mark the exact moments when solutions dropped from Q=Fix to Q=Float/Single or exhibited small position jumps despite retaining Fix status. This pattern appeared consistently across stations, confirming its systemic nature. A later series of five switches between 11:00 and 18:00 UTC produced similar degradation under continued ionospheric disturbance.

• Download figure
• Open in new tab
• Download PowerPoint

FIGURE 12

Frequent MGF switching during ionospheric disturbances on February 5, 2025: (a) dTEC map over Japan (Saito et al., 2014) from 00:00 to 06:00 UTC on February 5, 2025, with wave-like patterns indicating significant MSTID activity; (b) time series of MGF switching events (indicated by vertical dashed lines) and the corresponding kinematic positioning performance at representative stations

These frequent switches acted as a direct trigger for repeated loss of ambiguity resolution, preventing receivers from maintaining even a degraded fix that might otherwise have been possible under MSTID stress alone.

An important caveat applies to this web-archive-based analysis: corrections are provided in hourly files; thus, intra-hour switches are invisible in the data set. Consequently, real-time users likely experienced more frequent switching and correspondingly greater performance impacts than those observed here.

5.3.1 Reference-Frame Mismatch: System Offset in Static Vertical Component

The systemic vertical offset identified in Section 4.3.2 is primarily attributable to the reference-frame mismatch between CLAS (ITRF2014-based GSI F5) and IGS products (ITRF2020-based). This conclusion is corroborated by GSI’s own F5.1 trial analysis (GSI, 2025), which attributes significant vertical shifts to updated antenna models. This discrepancy is exacerbated by algorithmic differences between GSI’s baseline approach and CLAS’s state-space modeling, underscoring the importance of frame consistency.

This offset is not a transient anomaly but a chronic structural issue dating back to CLAS’s 2018 launch (then utilizing ITRF2005). This issue exemplifies the challenge that national-scale services face in reconciling rapidly evolving global standards (IGS) with stable national geodetic frameworks. This structural divergence creates a “cadence mismatch,” giving rise to distinct operational complexities:

• Inertia in Adoption: Large-scale infrastructure changes by global bodies such as the IGS can significantly impact many users, creating a barrier to rapid adoption.

• National vs. Global Cadence: As a national service, CLAS must align with Japan’s official geodetic framework (the GSI F5 solution), which is updated less frequently than global IGS products, creating an inevitable time lag.

Resolving this issue requires more than operational agility. SSR corrections effectively serve as real-time precise ephemeris products (Wübbena et al., 2005). Because of this equivalence, the reference frame of the SSR stream is inherently determined by the orbit and clock products used as its input backbone. Currently, CLAS relies on external products (typically aligned with the latest global IGS frame), which introduces the aforementioned frame mismatch against the national frame (GSI F5).

A potential solution is for CLAS to break this dependency and generate its own independent precise orbit and clock products. These products must be determined internally and be strictly consistent with the national geodetic framework, ensuring that the broadcast SSR is defined entirely within the target reference frame.

5.3.2 Reduced Number of Augmented Satellites: Causes and Implications

Section 4.3.1 identified a persistent decline in the number of augmented satellites in CLAS corrections throughout the study period. This trend comprises two distinct issues.

First, the long-term reduction primarily stems from the limited role of QZSS satellites in CLAS augmentation. The Cabinet Office is developing three additional satellites, targeting full seven-satellite constellation operations from JFY2026 (CAO, 2025a). This expansion is important, yet a structural challenge persists: L1-band signal fragmentation.

Although QZS-1R and subsequent Block III satellites introduce L1C/B, existing QZS-2 and QZS-4 lack it. The alternative common signal L1C, while present on all satellites, remains unsupported by most fielded receivers. The introduction of L1C/B—intended to maintain legacy navigation (LNAV) message compatibility during the transition from L1C/A—effectively served as a temporary measure to protect legacy receivers. However, within a modernization trajectory aiming for L1C/L5 unification, adding an interim signal while civil navigation (CNAV2) message-capable receivers remain scarce has exacerbated signal fragmentation. This issue complicates receiver implementation and delays the full benefits of modernization. Even after constellation expansion, the effective number of QZSS satellites available for high-accuracy augmentation may therefore remain constrained without widespread receiver upgrades.

This fragmentation reflects a structural misalignment—a cadence mismatch between the constellation’s modernization trajectory and the legacy receiver ecosystem’s upgrade cycles. A key consideration is the planned phase-out of the L2C signal (CAO, 2025a), which technically requires the existing user base—predominantly reliant on L1/L2 dual-frequency baselines—to migrate to a still-maturing L1C/L5 ecosystem. This timing mismatch between satellite modernization cycles and user equipment replacement cycles creates a transitional vulnerability window, during which the effective number of augmented satellites remains constrained, as observed in our data. In contrast to global trends, where multi-signal redundancy enhances robustness, the current transition period presents challenges for high-precision users until the receiver ecosystem catches up with constellation modernization.

Second, CLAS exhibits internal vulnerabilities causing anomalous satellite drops unrelated to QZSS outages (e.g., late 2023 and 2025 Q3). Moreover, the service has not kept pace with the rising number of theoretically visible multi-GNSS satellites, indicating operational constraints that prevent full resource utilization. Addressing these internal issues through improved operational processes may help mitigate such anomalies.

5.3.3 Impact of Frequent MGF Switching During Ionospheric Disturbances

Section 4.3.3 demonstrated how frequent MGF switching during severe ionospheric disturbances triggered repeated ambiguity resets and performance degradation—transforming a failsafe mechanism into a source of user-level instability.

This finding exposes two interconnected engineering trade-offs in the current CLAS ground architecture:

• System Isolation vs. Correction Consistency: Perfect state synchronization between MGFs would enable seamless switching but risks propagating a “poisoned” filter state (contaminated by unmodeled disturbances) to all systems, causing correlated total failure. Therefore, the present independent-state design deliberately prioritizes isolation—ensuring that at least one MGF can remain viable—over user continuity.

• Filter Responsiveness vs. Stability: Switching only to a “stable” standby is impractical when disturbances affect all MGFs simultaneously. More conservative filter tuning on standby systems (higher process noise, slower adaptation) could improve resilience but would sacrifice nominal accuracy.

Ultimately, frequent switching is not a simple flaw but a consequence of deliberate design choices that balance redundancy and robustness against extreme conditions, at the cost of user experience during stress events. The most effective mitigation lies in continuous refinement of the core generation algorithms themselves—making the algorithms inherently more resilient to disturbances that currently destabilize filters—combined with robust monitoring and validation logic.

5.3.4 Single Point of Failure: Dependence on GEONET Infrastructure

The multi-week correction outage in Compact Network ID 12 (Ogasawara) that occurred in March 2024 (Section 4.1) reveals a fourth vulnerability: a single point of failure arising from CLAS’s strict dependence on continuous GEONET availability. This dependency is not merely inferred; it is an officially acknowledged operational risk. The service provider (QSS) routinely issues Notice Advisory to QZSS Users (NAQU) warnings (e.g., NAQU2021028 (QSS, 2025d)) stating that “CLAS might be degraded or CLAS might be unusable” specifically “for the purpose of system maintenance of GEONET.”

Although PPP-RTK is promoted as reducing the need for local reference stations compared with conventional RTK, the bottom-up design of CLAS reintroduces a hard dependency on the underlying dense CORS network. As this event demonstrates, even planned or unplanned GEONET downtime—here due to prolonged station maintenance—can completely disable local atmospheric correction generation (subtype 12) for isolated regions, resulting in total service loss.

This vulnerability is particularly acute for geographically remote compact networks such as ID 12. To mitigate this issue, we recommend two complementary measures:

• Infrastructure Level: Incorporate backup CORS sources (e.g., private networks) for redundancy.

• Algorithmic Level: Implement graceful degradation—when local corrections become unavailable, the system should automatically fall back to global SIS corrections (standard PPP mode) rather than failing completely.

A more resilient architecture would adopt top-down-derived global SIS corrections as the foundational layer (Sato et al., 2024). This approach would not only prevent integration errors but also enable seamless graceful degradation to standard PPP when local atmospheric corrections are unavailable, ensuring continuous service availability and correction consistency even during infrastructure disruptions. Such evolution aligns with ongoing feasibility studies (Sato et al., 2024) and the recognition in national policy that CLAS generation algorithms require enhancement for broader resilience and scalability (CAO, 2025a).