{"id":"c84b1155-adc9-40db-8d5b-ee3cdd681250","arxiv_id":"2505.24077","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"A modular Python framework integrates real-time distributed acoustic sensing (DAS) data and machine-learning phase picks into the operational AQMS earthquake monitoring system, demonstrated on a 100-km Ridgecrest array.","lead":"The authors built a software framework that connects fiber-optic distributed acoustic sensing (DAS) data to the earthquake monitoring system used by the Southern California Seismic Network. It streams selected DAS channels and injects machine-learning traveltime picks into the operational AQMS workflow, demonstrated on a 100-km array in Ridgecrest, California.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The >400 aftershock 'detection' is unvalidated and based on all 5,000 channels, not the 18-channel AQMS subset, leaving the operational integration claim quantitatively unsupported.","rationale":"The central claim packages two things: a software integration architecture and a demonstration that the resulting system provides value for operational monitoring. The paper provides strong evidence for the software plumbing (telemetry reliability, packet handling) and for the existence of picks, but the value proposition rests on the 400-aftershock claim. That claim is not quantified against a reference catalog and, crucially, is produced by processing all 5,000 streamed channels with PhaseNet-DAS, whereas the AQMS integration only ingests picks from 18 channels (8 in operations). A reader cannot tell whether the integrated subset can detect even a fraction of those aftershocks. The acknowledged lack of a phase-association step further means that mispicks like those in Figure 3b would enter the AQMS pick stream. This is the weakest link in the argument: if the 18-channel detection rate is low or false alarms are high, the framework may integrate technically without providing operational value. The proposed test isolates this by comparing the same pipeline restricted to the integrated subset against the full-array performance and a reference catalog, using the released code and metadata. The reader's verdict of CONDITIONAL is appropriate; this concern reinforces it rather than changing it.","tokens_in":12804,"tokens_out":8707,"duration_ms":89257,"concrete_test":"Reprocess the archived Ridgecrest DAS data for August 8-30, 2024 using StreamProcessor.py with the 18-channel XML metadata and the 70-s rolling PhaseNet-DAS picking, and feed only the 18 channels' picks to the pick ring. Associate picks to the SCSN/USGS catalog within 150 km using predicted P/S arrival windows of ±5 s and a simple ≥3-channel event association. Report the true detection rate, false-alarm rate, and median pick residual for the 18-channel subset and for the full 5,000-channel picking. If the 18-channel detection rate is below 50% of the full-array rate or false alarms exceed one per hour, the operational integration claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative evidence for operational value is the claim that the real-time PhaseNet-DAS workflow 'generated more than 400 aftershocks' of the M5.2 Lamont sequence (Section 'Real-time processing and earthquake sequence example', third paragraph). This count is never defined or compared against the SCSN/USGS catalog, so genuine detections cannot be distinguished from false picks. Moreover, the illustrative data in Figure 5b are picks from all 5,000 streamed channels, whereas the AQMS integration only receives picks from the 18 selected channels (currently 8 in operations). The number of Lamont aftershocks detectable using only those 18 channels is not reported, and no AQMS-generated event list is presented. The M1.0 example in Figure 3b shows misclassified P arrivals, confirming that pick quality degrades for small events; the paper even acknowledges that a phase-association step is not included in the software, so such mispicks would be passed to AQMS. Without quantitative detection rate, precision/recall, and latency metrics for the actual integrated channel subset, the claim that this framework enables DAS to contribute to operational earthquake monitoring rests on an unvalidated assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a modular Python software framework for integrating distributed acoustic sensing (DAS) data into the ANSS Quake Monitoring Software (AQMS) used by the Southern California Seismic Network. The system streams packets from a Ridgecrest, California DAS array into 1-second Earthworm wave rings, applies the PhaseNet-DAS machine-learning picker to a rolling 70-second buffer, and injects traveltime picks from 18 selected DAS channels into an Earthworm pick ring for ingestion by the AQMS associator. The demonstration claims continuous streaming since August 2024 with telemetry reliability exceeding 99.989% and detection of more than 400 aftershocks of the M5.2 Lamont sequence. The paper also provides implementation details, command-line usage examples, and a discussion of future amplitude-based capabilities.","tokens_in":13097,"tokens_out":5665,"duration_ms":57164,"significance":"If the claimed performance holds, this is a valuable engineering contribution: it is one of the first documented, open-source pipelines for passing DAS-derived picks into an operational seismic network's AQMS workflow. The paper's strengths include a detailed architecture description, use of standard SEED metadata, reproducible command examples, an accessible code repository, and a real-data deployment on an 80-km effective-length array. The central operational claim, however, rests on validation that is currently qualitative and partly performed on a different channel set from the one actually integrated into AQMS, so the quantitative impact on monitoring operations is not yet established.","major_comments":[{"comment":"The statement that the real-time PhaseNet-DAS workflow 'generated more than 400 aftershocks' of the M5.2 Lamont sequence is not quantitatively validated. No reference-catalog comparison, detection-rate, precision/recall, or false-positive statistics are reported, and the assertion that the picks are 'in good agreement with the detections from nearby SCSN stations' is qualitative. Because this count is the principal evidence of operational value, it must be supported by a defined metric and a direct comparison against the SCSN/USGS catalog.","section":"Real-time processing and earthquake sequence example"},{"comment":"The AQMS integration is demonstrated with picks from only 18 selected channels, yet the 400+ aftershock example (Figure 5b) uses PhaseNet-DAS picks from all 5,000 streamed channels. The number of aftershocks detectable from the 18-channel subset—the configuration actually inserted into the AQMS pick ring—is not reported, and no AQMS-generated event list is presented. The central claim that this framework enables DAS to contribute to operational monitoring therefore applies to a configuration that is not the one quantitatively demonstrated.","section":"Real-time processing and earthquake sequence example"},{"comment":"The M1.0 example in Figure 3b contains misclassified P arrivals, and the text states that no phase-association step is included in the software. Since these picks are streamed directly to the Earthworm pick ring and hence to the AQMS associator, the paper should either quantify the expected false-pick rate for small events or add a preprocessing phase-association filter (as suggested via the cited Zhu et al. 2022 work) before claiming operational suitability.","section":"Real-time processing and earthquake sequence example"},{"comment":"The 'real-time' claim lacks an end-to-end latency measure. The reported 0.6-second packet latency and approximately 1 second per picking task do not bound the delay between an earthquake's occurrence and the insertion of picks into the AQMS pick ring, because the 70-second rolling buffer introduces a window-dependent delay. Please report end-to-end detection-to-ingestion latency, for example by comparing pick timestamps with catalog origin times for the Lamont sequence.","section":"Real-time DAS data streaming and processing"}],"minor_comments":[{"comment":"The term 'wavering' is used repeatedly (e.g., 'inserted into an Earthworm wavering' and the '-wring' option) and appears to be a typo for 'wave ring' or the Earthworm ring name; please correct and define the intended Earthworm module.","section":"Code description and Usage examples"},{"comment":"In the command-line examples, the description for --strnRt reads 'Compute realt-time strain rate from strain'; fix the typo and explain what the flag computes.","section":"Usage examples"},{"comment":"Figure 4 is based on events from May-July 2023, before the continuous streaming operation that began August 2024; clarify that this SNR analysis used archived data and is not part of the real-time pipeline.","section":"Real-time processing and earthquake sequence example"},{"comment":"The text says 'The instrument transmits individual data packets at 100 Hz' after stating that the ping rate is 1 kHz and data are stored at 100 Hz; it would be clearer to distinguish the packet transmission rate from the decimated sample rate.","section":"Data streaming and selected channel metadata"},{"comment":"The 5-km channel-spacing rule is described as chosen to be 'comparable with SCSN regions with dense station coverage'; a citation or quantitative justification for this value would help readers assess the trade-off between coverage and redundancy.","section":"Real-time DAS data streaming and processing"}],"recommendation":"major_revision","confidential_remarks":"This is a software and engineering contribution whose scope fits a seismological methods journal. The gaps identified are addressable with additional analysis rather than fundamental errors; in particular, adding a catalog-based evaluation for the 18-channel AQMS subset would greatly strengthen the operational claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nWhat you should know: this is a solid engineering paper about integrating DAS into a real seismic network (SCSN/AQMS). It's not a breakthrough in DAS physics or earthquake science, but it is one of the first operational integrations of its kind, and the software is real, shipped, and running continuously since August 2024. The modular design—StreamProcessor, RingBuffer, SEED metadata for selected channels, PhaseNet-DAS picks into Earthworm pick rings—is clear and reusable. Telemetry reliability >99.989% and 0.6-second latency per packet give hard evidence that the plumbing works.\n\nThe paper does what it sets out to do: it shows a working path to feed DAS waveforms and ML picks into AQMS without breaking the existing workflow. The M3.4 example shows clean picks; the M1.0 example is honestly presented with misclassified P-arrivals. The authors also explicitly state that a phase-association step is not included, which is a real limitation they acknowledge.\n\nNow the soft spots, which are real but not fatal. The headline number—\"more than 400 aftershocks\" of the Lamont M5.2 sequence—is not validated. There's no comparison to the SCSN/USGS catalog, no precision/recall, no false-positive rate, and, as the stress-test notes, those picks come from all 5,000 streamed channels, not from the 18 selected channels that feed AQMS (8 currently in operations). So this number doesn't demonstrate that the operational integration improves earthquake detection. Similarly, there are no end-to-end latency figures for pick-to-association, and no AQMS-generated event list from the DAS picks. The paper's central claim is about enabling integration, which is plausible and demonstrated qualitatively; the quantitative operational benefit remains an assertion. The discussion of strain-to-ground-motion conversion is appropriately cautious.\n\nThe citation pattern looks fine; PhaseNet-DAS and other DAS works are cited, and the authors don't oversell novelty. The free parameters (channel spacing, buffer length, dedup window) are operational choices, not fitted to make a result. So no circularity concern.\n\nWho is this for? Seismic network operators, DAS software developers, and anyone planning to interface DAS with AQMS or similar systems. It deserves a serious referee; with a quantitative evaluation section added (catalog comparison for the 18-channel subset, pick latency, false-positive counts), it would be a stronger paper. I'd accept it for review in its current form, but I'd push for that validation.\n\nRecommendation: send to peer review. Ask the authors to add quantitative detection statistics for the actual AQMS-ingested channels.","headline":"A real, well-engineered DAS-to-AQMS integration that deserves review, but its operational benefit is asserted more than demonstrated.","tokens_in":13597,"tokens_out":2276,"would_cite":true,"duration_ms":23707,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Real-time DAS data can be folded into an operating earthquake monitoring system by treating selected fiber channels as standard seismic stations and feeding machine-learning phase picks into the existing AQMS pipeline.","keywords":["distributed acoustic sensing","real-time earthquake monitoring","AQMS","PhaseNet-DAS","Earthworm","traveltime picking","Ridgecrest DAS array","fiber-optic seismology"],"falsifier":"Compare every PhaseNet-DAS pick streamed into AQMS during a fixed one-month window against a reference catalog for events within 120 km of the Ridgecrest array; if the live pick stream misses a large share of catalog events at local magnitude 2 and above, or produces false associations that the associator cannot reject, the operational-accuracy assumption is refuted.","tokens_in":12680,"feed_emoji":"📡","tokens_out":10166,"duration_ms":86694,"temperature":0.7,"pith_summary":"The paper argues that distributed acoustic sensing (DAS) data can be folded into an operating earthquake monitoring center by treating a handful of fiber-optic channels as ordinary seismic stations and letting the deep-neural-network picker PhaseNet-DAS supply the phase arrivals. It presents a modular Python framework that streams DAS packets into 1-second Earthworm ring buffers under SEED channel metadata, while PhaseNet-DAS runs asynchronously on a rolling 70-second window and injects its traveltime picks into the AQMS associator. The demonstration is continuous: since August 2024 the 80-km Ridgecrest array has streamed 5,000 channels to a processing server with 0.6-second packet latency and better than 99.989% telemetry reliability, and the real-time pipeline detected more than 400 aftershocks of the August 2024 M5.2 Lamont sequence. If the approach holds in routine operations, monitoring networks can supplement sparse seismometer coverage with dense fiber-optic arrays without replacing their existing software stack.","feed_headline":"Live fiber-optic sensing captures 400+ aftershocks","feed_subtitle":"A modular pipeline streams dense DAS data into standard quake-monitoring software with machine-learning picks.","key_machinery":"The carrying object is a translation-and-buffer pipeline: a StreamReader template that unpacks vendor DAS packets (demonstrated with an OptaSense interrogator), a RingBuffer that accumulates samples into 1-second Earthworm wavering rings and separately into a 70-second working window, and an asynchronous PhaseNet-DAS inference loop that sends picks to an Earthworm pick ring through the PyEarthWorm interface. SEED metadata definitions such as CI.DRS02..HS1 let existing AQMS tools treat a DAS channel as a standard station, and PhaseNet-DAS, a deep neural network trained on DAS data for P/S arrival-time picking, is the component that exploits dense spatial sampling. The sparsity choice of selecting channels every ~5 km keeps the DAS input commensurate with conventional station density, while the full 5,000-channel stream is still used for picking.","core_discovery":"The central claim is that an operational earthquake monitoring system such as AQMS can ingest real-time DAS data with no change to its internal logic, provided a translation layer buffers DAS samples into standard 1-second traces and only a sparse subset of well-coupled channels (roughly 5 km spacing) is forwarded. On top of that, the full DAS array can be exploited by running PhaseNet-DAS continuously on a rolling 70-second buffer over all streamed channels; picks from the selected channels are filtered to remove duplicates within 1 second and sent to an Earthworm pick ring for phase association. The paper reports this has been running since August 2024 on the Ridgecrest array, with 18 selected channels feeding AQMS and 8 already used in network monitoring operations. For events recorded between May and July 2023, pick signal-to-noise ratios increase with magnitude, and for the M5.2 Lamont sequence the real-time picks agree with nearby network detections while STA/LTA thresholding on the same DAS data misses many aftershocks and adds traffic-related mispicks. The system therefore claims to make DAS a practical supplement to conventional stations, not just a research instrument.","pith_inferences":["A natural extension the paper leaves implicit is feeding the dense picks from all 5,000 channels into a phase-association step rather than only the 18 selected channels, which could substantially improve event location and detection completeness.","The authors do not report end-to-end alert latency; if packet latency (~0.6 s) and picking time (~1 s) hold, the design sits near the threshold for earthquake early warning, but a quantified alert-latency budget would be needed to test that.","If the DAS amplitude calibration questions referenced in the paper are resolved, the same pick ring could carry strain-rate amplitudes, enabling real-time magnitude estimation directly from fiber-optic data without adding a new waveform path.","The framework's modularity implies it could be transferred to submarine cables or urban telecom networks, but transferability depends on the pre-trained PhaseNet-DAS generalizing to those installations, which is not established here."],"forward_implications":["Regional networks using AQMS can add DAS channels without rewriting associators or archive tools, since selected channels enter as standard SEED traces in Earthworm rings.","The demonstrated telemetry reliability (over 99.989% since August 2024) suggests dark-fiber arrays can support continuous operational data feeds, not just campaign experiments.","Real-time PhaseNet-DAS picks on the full array detect small aftershocks that STA/LTA thresholding on the same DAS data misses, pointing to better catalog completeness at low magnitudes.","Because picking runs asynchronously on a rolling buffer, the design can absorb faster or slower inference models and larger channel counts by adjusting buffer length and worker concurrency.","Archiving the DAS strain-rate waveforms lets analysts review and manually pick events in the same Jiggle interface used for conventional stations."],"supporting_citations":[{"why":"Supplies the PhaseNet-DAS deep-learning picker that the real-time pipeline runs; the paper's detection results depend on its pretrained weights.","marker":"W. Zhu et al., 2023"},{"why":"Describes the AQMS open-source monitoring software whose Earthworm rings and pick ingestion the framework targets.","marker":"Renate Hartog et al., 2020"},{"why":"Provides PyEarthWorm, the Python interface used to send PhaseNet-DAS picks into the Earthworm pick ring.","marker":"Hernandez and Martinez-Torres (2018)"},{"why":"Provides the vehicle-based geolocation method that assigns coordinates to the Ridgecrest DAS channels used in the SEED metadata and channel selection.","marker":"Biondi, Wang, et al. (2023)"},{"why":"Shows that a shorter Ridgecrest DAS array increased detected aftershocks of the 2019 M7.1 event by two orders of magnitude, motivating the real-time monitoring claim.","marker":"Z. Li et al., 2021"},{"why":"Defines the STA/LTA detector that the paper compares against PhaseNet-DAS on the Lamont sequence to show missed aftershocks and traffic mispicks.","marker":"Trnkoczy, 2009"},{"why":"Supports the flat-instrument-response assumption used to convert raw DAS phase values to microstrain rate for archiving and picking.","marker":"Paitz et al., 2021"}],"fun_headline_variants":["Real-time DAS now streams into standard quake monitoring ops","DAS data integrated into operational earthquake monitoring with ML picks","Fiber-optic array feeds live picks into AQMS without system changes","Seamless DAS streaming into ANSS via standardized seismic formats","Ridgecrest DAS array goes live in network monitoring since 2024"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The pre-trained PhaseNet-DAS model, applied to Ridgecrest data without fine-tuning, is accurate enough for real-time operational picking; the paper shows only qualitative examples and does not quantify detection rates, false positives, or pick latency against a reference catalog.","fun_headline_variants_meta":{"raw":{"variants":["Real-time DAS now streams into standard quake monitoring ops","DAS data integrated into operational earthquake monitoring with ML picks","Fiber-optic array feeds live picks into AQMS without system changes","Seamless DAS streaming into ANSS via standardized seismic formats","Ridgecrest DAS array goes live in network monitoring since 2024"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000297,"raw_usage":{"total_tokens":1715,"prompt_tokens":933,"completion_tokens":782,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":689}},"tokens_in":549,"tokens_out":782,"duration_ms":7832,"temperature":1.0,"reasoning_tokens":689,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:35:01.946867+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare every PhaseNet-DAS pick streamed into AQMS during a fixed one-month window against a reference catalog for events within 120 km of the Ridgecrest array; if the live pick stream misses a large share of catalog events at local magnitude 2 and above, or produces false associations that the associator cannot reject, the operational-accuracy assumption is refuted.","supporting_citations":[{"cited_title":", Friberg, P A","cited_arxiv_id":null,"evidence_quote":"Describes the AQMS open-source monitoring software whose Earthworm rings and pick ingestion the framework targets."},{"cited_title":"\\ Martinez-Torres, F A","cited_arxiv_id":null,"evidence_quote":"Provides PyEarthWorm, the Python interface used to send PhaseNet-DAS picks into the Earthworm pick ring."},{"cited_title":"APACrefauthors \\ 2009","cited_arxiv_id":null,"evidence_quote":"Defines the STA/LTA detector that the paper compares against PhaseNet-DAS on the Lamont sequence to show missed aftershocks and traffic mispicks."},{"cited_title":", Edme, P","cited_arxiv_id":null,"evidence_quote":"Supports the flat-instrument-response assumption used to convert raw DAS phase values to microstrain rate for archiving and picking."}],"review_version":1}