{"id":"b4b2f75d-13e6-42f9-b7f1-ade801f46f56","arxiv_id":"2505.04772","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Starlink's flat high performance terminal in Oulu, Finland loses roughly half its uplink and over a third of its downlink throughput in rain, cloud cover correlates with lower throughput, and round-trip time is unaffected.","lead":"This paper measures how rain and cloud cover affect Starlink's flat high performance terminal in Oulu, Finland, finding that rain cuts median uplink and downlink throughput by about 52% and 38%, while latency stays flat. It offers one of the first weather-performance datasets for Starlink's recommended Arctic terminal, useful for planning satellite links in northern regions.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The cloud-cover regression is the least secure pillar: one 148-minute session, remote airport cloud data, and no reported regression statistics make the abstract's cloud claim non-reproducible as presented.","rationale":"The paper is a measurement study whose central quantitative claims are rain-driven throughput degradation and a cloud-cover trend. The rain claim is comparatively well supported: the VTT weather station is close to the client, there are multiple measurement days, and the observed degradation is large and consistent with prior Ku-band satellite literature. Its main weakness is an internal numerical inconsistency (abstract: 37.84% downlink; Section V-A: 35% downlink), which is a reproducibility issue but not a threat to the qualitative conclusion. The cloud-cover claim, however, is the load-bearing weak point because the entire analysis rests on a single day, a remotely located cloud-cover source, and a regression whose outputs are never reported. In Fig. 9 the cloud cover monotonically increases over the 148-minute session, so the decreasing throughput in Fig. 12 could equally be explained by time-of-day effects, changing satellite geometry, or network congestion. Without regression coefficients, confidence intervals, or a time-controlled robustness check, the abstract's quantitative cloud claim cannot be independently evaluated. The manuscript itself acknowledges the need for further measurements in the conclusion, which supports a conditional rather than a definitive verdict. The proposed test would settle whether the cloud effect survives time as a control; if it does not, the abstract should be revised to remove or substantially weaken the cloud claim. This concern is the same one the reader identified, so the reader's CONDITIONAL verdict stands unchanged.","tokens_in":9824,"tokens_out":6628,"duration_ms":68598,"concrete_test":"Re-analyze the raw June 26 per-second throughput and cloud-cover series (requesting data from the authors if needed) by fitting the Section V-B linear regression with throughput as response and cloud cover as predictor, and report slope, standard error, R², and p-value. Then add a time index or session-half fixed effect to the same model. If the cloud-cover coefficient is not significant when time is controlled, or if R² is negligible, the negative cloud relationship is a temporal artifact rather than a weather effect.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The cloud-cover result in Section V-B is the least secure pillar of the central claims. It rests on one 148-minute session (June 26, 2024), on cloud cover measured at FMI Oulu Airport about 15 km from the terminal (Section IV-A), and on an unspecified linear regression with no reported coefficients, R², or p-values. The cloud cover rises monotonically from near 0% to 87.5% over the session (Fig. 9), so any decreasing throughput trend (Fig. 12) is confounded by time, satellite pass geometry, and network conditions. The 25 km spot-beam justification is also not sufficient because the relevant quantity for Ku-band attenuation is cloud water along the slanted terminal-to-satellite path, not ground-level okta at a remote airport. Since the abstract states a quantitative cloud comparison ('about 20% higher throughput at <=12.5% vs 87.5% cover'), the lack of regression diagnostics and the single-day design make this headline claim non-reproducible as presented. The rain-degradation claim is better supported by local weather data and multiple days, though the abstract's 37.84% downlink figure conflicts with the 35% stated in Section V-A; this is secondary.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a measurement study of a Starlink flat high performance (FHP) terminal located in Oulu, Finland, near the Arctic Circle. Using iPerf3, ping, and traceroute measurements collected on four days in June 2024, the authors compare TCP uplink and downlink throughput and round-trip time under clear, rainy, and cloudy conditions. The central claims are that rain degrades median uplink and downlink throughput by about 52% and 35--38%, respectively, that RTT is unaffected by rain and clouds, and that linear regression of throughput against cloud cover reveals a negative relationship, with roughly 20% higher throughput at cloud cover up to 12.5% than at 87.5% cover. The paper also documents one-second outages during moderate rain and reports service availability above 98.5%.","tokens_in":10117,"tokens_out":2585,"duration_ms":26089,"significance":"The study fills a gap in the literature by evaluating the Starlink FHP terminal at high latitudes using local weather-station data, and it provides a bidirectional TCP throughput and RTT comparison across weather conditions. The rain-degradation result is qualitatively supported by multiple measurement days and local rain-intensity data, and the RTT finding is consistent with prior work. The cloud-cover result is the weakest pillar: it rests on one measurement session, remote airport cloud data, and an undescribed regression. If the authors supply regression diagnostics and address confounds, the paper would be a useful empirical contribution; the weather-related quantitative claims are currently not reproducible as presented.","major_comments":[{"comment":"The abstract reports a median downlink degradation of 37.84%, while Section V-A and Fig. 6 state 35%; the uplink figures also differ slightly (52.27% vs. 52.2%). Because these percentages are headline quantitative claims, the authors must specify which data subsets and which baseline values are used for each percentage and make the numbers internally consistent.","section":"Abstract and Section V-A"},{"comment":"The cloud-cover analysis rests on a single 148-minute session on June 26, 2024, and the described linear regression is not accompanied by a slope, intercept, R^2, confidence intervals, or p-values. Since cloud cover rises monotonically from near 0% to 87.5% over the session (Fig. 9), the decreasing throughput trend in Fig. 12 is confounded with time, satellite pass geometry, and network conditions. The abstract's claim of approximately 20% higher throughput at low cloud cover is therefore non-reproducible as presented; the authors should report full regression statistics and address time-based confounding, for example by controlling for time of day or collecting additional sessions.","section":"Section V-B"},{"comment":"The assumption that cloud cover measured at FMI Oulu Airport, about 15 km from the client, represents conditions at the Starlink terminal is justified only by an unspecified 25 km spot-beam diameter and an unknown elevation angle. Ku-band attenuation depends on the cloud water content along the slanted terminal-to-satellite path, not on ground-level okta at a remote airport, so this assumption is not sufficient. The authors should provide empirical validation (e.g., comparison with co-located cloud observations) or explicitly downgrade the cloud-cover analysis to a preliminary observation.","section":"Section IV-A"}],"minor_comments":[{"comment":"The caption describes an upper green plot and a lower red plot, but the color and position descriptions appear inconsistent with the displayed panels; please clarify which panel corresponds to June 18 and which to June 27.","section":"Fig. 4 caption"},{"comment":"The sentence 'Unlike iperf3 throughput, we separately examine the RTT for these two servers' is missing a period before the following sentence beginning 'We conducted...'.","section":"Section IV-B"},{"comment":"Minor typographical issues include 'ten minutes internals' and 'drizzing'; the abstract also contains 'clouds cover' in places. A proofreading pass is recommended.","section":"Section V-A"},{"comment":"The timing diagram in Fig. 11 should be reconciled with the reported sample counts (3,690 uplink, 3,690 downlink, and 492 RTT samples) so that readers can verify that the repetitions and durations are consistent.","section":"Section V-B and Fig. 11"},{"comment":"The conclusion's statement that moderate to heavy rainfall may contribute to higher RTT is presented as an expectation rather than a measured result; please mark it clearly as a hypothesis for future work.","section":"Section VI"}],"recommendation":"major_revision","confidential_remarks":"This is a legitimate measurement paper with a sound rain-related core, but the cloud-cover headline claim is not reproducible as presented due to the single-session design, remote cloud data, and missing regression statistics. If the authors can add regression diagnostics and reframe the cloud result as preliminary, or strengthen it with additional data, the paper could become acceptable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful measurement paper, not a breakthrough. It adds the first bidirectional Starlink FHP terminal weather data from near the Arctic Circle. The rain analysis is mostly solid; the cloud analysis is not. Fix the inconsistencies in the degradation percentages and either report full regression statistics or drop the quantitative cloud claim from the abstract.\n\nWhat's new: previous weather studies used other terminals or regions; this one uses the FHP terminal in Oulu, Finland, with throughput and RTT both directions, multiple iPerf servers, and a local weather station a few meters from the vehicle for rain/humidity/temperature. The finding that rain degrades uplink and downlink but leaves RTT essentially unchanged is consistent with the literature, and the data in Figs. 5-8 support it. The service availability result (above 98.5% even in rain) is a useful data point. The paper is honest that moderate-to-heavy rain is under-tested.\n\nSoft spots: the abstract says downlink degradation is 37.84%, but Section V-A says 35%, and Fig. 5's own medians imply about 34%. Pick one and trace it. The cloud-cover regression is the bigger problem. It rests on one 148-minute session on June 26, cloud cover from FMI Oulu Airport about 15 km away, and no reported slope, R^2, confidence interval, or p-value. Cloud cover rises monotonically over the session, so the negative trend is confounded with time of day and satellite pass geometry. The spot-beam diameter argument doesn't address the actual physical quantity (cloud water along the slant path). The 20% comparison between 12.5% and 87.5% coverage is essentially comparing two endpoints of a single time series. That claim should either be backed with proper regression diagnostics and ideally another day, or removed from the abstract. The rain claims, by contrast, use multiple days and local weather stations, so they carry more weight.\n\nWho should read it: people planning 6G NTN deployments or Starlink use in Nordic conditions, and the Starlink measurement community. With the numeric inconsistencies fixed and the cloud analysis strengthened or softened, it's a reasonable conference paper. As it stands, it's borderline but worthy of a serious referee, mainly because the FHP/Northern Europe rain data is genuinely new.\n\nI'd bring it to reading group if we're doing satellite measurements; otherwise probably not. I'd cite it only after the numbers are cleaned up.","headline":"First FHP terminal weather data near the Arctic, useful rain results, but the cloud-cover headline and abstract percentages need fixing before this is trustworthy.","tokens_in":10600,"tokens_out":2771,"would_cite":false,"duration_ms":25288,"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":"Rain cuts Starlink median uplink throughput by 52% and downlink by 38%, while latency stays flat; cloud cover adds a further ~20% throughput penalty.","keywords":["Starlink","LEO satellite communications","weather impact","rain attenuation","cloud cover","throughput measurement","round-trip time","Ku-band"],"falsifier":"Repeat the cloud-cover comparison with a sky-facing camera or local cloud sensor mounted at the Starlink terminal over a month of variable cloudiness; if throughput does not decline with locally measured cloud fraction, the airport-based cloud regression is an artifact of distance or of the one afternoon. For the rain claim, bin the rain samples by rain rate at the terminal: if median uplink throughput does not drop by roughly half across light-to-moderate rain events, the headline 52% figure does not reproduce.","tokens_in":9678,"feed_emoji":"🌧️","tokens_out":10928,"duration_ms":102539,"temperature":0.7,"pith_summary":"Low-Earth-orbit satellite internet is promoted as weather-tolerant, but this paper measures how far that tolerance goes for Starlink's flat high performance terminal, a fixed phased-array antenna rated for extreme weather, near the Arctic Circle. On five measurement days in Oulu, Finland, rain lowered median uplink throughput by 52.27% and median downlink throughput by 37.84%, while round-trip time stayed effectively unchanged. A linear regression over cloud-cover data shows throughput falling as cloud cover rises: skies up to 12.5% cloudy delivered roughly 20% higher throughput than 87.5% cloudy skies. The authors conclude that weather degrades Ku-band Starlink throughput substantially without breaking connectivity, since service availability stayed above 98.5% on both links during rain.","feed_headline":"Rain cuts Starlink uplink throughput by half; cloud cover adds 20%","feed_subtitle":"Five-day Arctic measurements show latency holds steady even as rain and clouds shrink median throughput.","key_machinery":"The carrying object is a measurement campaign centered on Starlink's flat high performance (FHP) terminal, a fixed-installation phased-array antenna rated for extreme weather, mounted on a stationary vehicle. Throughput was measured with iPerf3 over TCP using ten parallel streams in both directions, RTT with ping, and weather with local stations reporting rain intensity, humidity, temperature, and cloud cover at one-minute resolution. The analysis pools results across five server locations because earlier work with the same hardware showed server distance had no material effect, and then compares weather classes or regresses throughput against cloud cover. This design lets the authors separate weather effects from network-path effects in a single geographic setting.","core_discovery":"The central claim is that weather attenuates Starlink's radio link asymmetrically. In rain, median uplink throughput falls from 20.9 to 10.5 Mbps (a 52.27% drop) and median downlink from 137 to 90.2 Mbps (a 37.84% drop), while median RTT remains essentially the same; moderate rain also produced a handful of one-second uplink outages. The cloud analysis adds a negative linear relationship between cloud fraction and throughput, with about 20% more throughput at up to 12.5% cloud cover than at 87.5%, and no visible RTT trend. The paper attributes these losses to Ku-band rain and cloud attenuation, and notes that most rain samples were light rain, so heavier rain would likely push throughput lower.","pith_inferences":["Beyond the paper, the same data could test whether rain attenuation scales with elevation angle: at low angles the signal crosses more storm volume, so the 52% uplink drop should be larger when the satellite sits near the horizon, a check possible from satellite ephemeris data.","Beyond the paper, the cloud slope was measured on a single afternoon and is correlated with rising humidity, so a longer campaign with a local sky-facing sensor could separate cloud attenuation from humidity effects.","Beyond the paper, if this behavior holds across Ku-band LEO providers, weather-aware traffic shaping—pre-fetching during clear periods and throttling during rain—could smooth user experience without new infrastructure."],"forward_implications":["Users of Ku-band LEO services in rainy high-latitude regions should expect roughly a 40-50% throughput reduction during rain even with a premium weather-resistant terminal.","Because round-trip time stays flat through rain and clouds, latency-sensitive applications can continue during weather, but large transfers will slow unless a fallback link exists.","Service availability above 98.5% during the measured rain indicates that light-to-moderate rain is a soft throughput degradation rather than a hard outage.","A linear cloud-cover penalty of about 20% between nearly clear and nearly overcast skies means overcast weather alone meaningfully lowers achievable throughput.","Link budgets for future Ku-band LEO constellations should treat cloud cover as a continuous attenuation factor rather than a binary clear/rain condition."],"supporting_citations":[{"why":"It supplies the earlier end-user Starlink measurement showing precipitation cuts throughput by about 27% on average, the baseline this paper's FHP-terminal result extends.","marker":"[6]"},{"why":"It provides the hardware setup details and the server-location finding that this campaign reuses, allowing throughput results to be pooled across servers.","marker":"[15]"},{"why":"It reports the Germany/Netherlands rain result of roughly 30% UDP throughput loss at 4-5 mm/h rain, a direct comparison for the measured degradation.","marker":"[20]"},{"why":"It documents in Sweden that Starlink throughput decreases with precipitation, the closest regional precedent for the Oulu rain results.","marker":"[22]"},{"why":"It reports reduced UDP download throughput under cloudy conditions and a stable RTT, which the cloud-cover analysis aligns with and extends.","marker":"[23]"}],"fun_headline_variants":["Rain slashes Starlink uplink 52%, downlink 38%","Arctic rain halves Starlink uplink, RTT steady","Cloud cover shaves 20% off Starlink throughput","Rain cuts Starlink speeds, ping unaffected"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The cloud-cover result stands or falls on whether cloud observations taken at an airport roughly 15 km from the terminal truly represented the sky above the terminal on that single measurement day.","fun_headline_variants_meta":{"raw":{"variants":["Rain slashes Starlink uplink 52%, downlink 38%","Arctic rain halves Starlink uplink, RTT steady","Cloud cover shaves 20% off Starlink throughput","Rain cuts Starlink speeds, ping unaffected"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00026,"raw_usage":{"total_tokens":1604,"prompt_tokens":974,"completion_tokens":630,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":590,"completion_tokens_details":{"reasoning_tokens":557}},"tokens_in":590,"tokens_out":630,"duration_ms":6116,"temperature":1.0,"reasoning_tokens":557,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:21:03.997255+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the cloud-cover comparison with a sky-facing camera or local cloud sensor mounted at the Starlink terminal over a month of variable cloudiness; if throughput does not decline with locally measured cloud fraction, the airport-based cloud regression is an artifact of distance or of the one afternoon. For the rain claim, bin the rain samples by rain rate at the terminal: if median uplink throughput does not drop by roughly half across light-to-moderate rain events, the headline 52% figure does not reproduce.","supporting_citations":[{"cited_title":"Network characteristics of LEO satellite constellations: A starlink-based measurement from end users,","cited_arxiv_id":null,"evidence_quote":"It supplies the earlier end-user Starlink measurement showing precipitation cuts throughput by about 27% on average, the baseline this paper's FHP-terminal result extends."},{"cited_title":"WetLinks: a large-scale longitudinal Starlink dataset with contiguous weather data,","cited_arxiv_id":null,"evidence_quote":"It reports the Germany/Netherlands rain result of roughly 30% UDP throughput loss at 4-5 mm/h rain, a direct comparison for the measured degradation."},{"cited_title":"Throughput analysis of Starlink satellite internet: A study on the effects of precipitation and hourly variability with TCP and UDP,","cited_arxiv_id":null,"evidence_quote":"It documents in Sweden that Starlink throughput decreases with precipitation, the closest regional precedent for the Oulu rain results."},{"cited_title":"Observing the skies - ground-based cloud detection for evaluating the impact of clouds on leo communications,","cited_arxiv_id":null,"evidence_quote":"It reports reduced UDP download throughput under cloudy conditions and a stable RTT, which the cloud-cover analysis aligns with and extends."}],"review_version":1}