{"id":"462e7832-93d1-4fd2-aa6b-63e1e364f397","arxiv_id":"2504.16878","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Using mobility data from six recent U.S. sequential tropical cyclone pairs, the study finds higher preparedness for the second storm, diminishing sensitivity to forecast wind, and power-outage spillover effects on neighboring counties.","lead":"This paper measures how people prepared for back-to-back hurricanes in six U.S. cases from 2020 to 2024, using cellphone-mobility data on visits to gas stations, grocery stores, and building-supply shops. It finds that people prepare more for the second storm but respond less to its forecast strength, and that power outages from the first storm drive higher preparedness and spill over to neighboring counties.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Baseline defined before first storm likely inflates second-TC preparedness; needs post-first-storm baseline robustness test.","rationale":"The reader's weakest-assumption diagnosis identifies exactly the same structural risk: the baseline is set before the first storm, so the second-TC preparedness ratio can reflect post-disaster activity rather than anticipatory preparedness for the second storm. This is the most load-bearing concern because it threatens the headline claim about higher preparedness for the second TC, and it also interacts with the forecast-wind interaction term. The concern is concrete, located in Methods, and testable. I do not see a more fundamental issue that would change the conditional verdict. The paper otherwise uses real mobility data, transparent regression specifications, and multiple robustness checks; the lack of code/data and the Helene-Milton exclusion are secondary. A clean acceptance would require the proposed baseline robustness check, so the reader's CONDITIONAL verdict remains appropriate.","tokens_in":14448,"tokens_out":2890,"duration_ms":29423,"concrete_test":"Re-estimate Table 1 Model 2 (and the logistic robustness model) using an alternative baseline for the second-TC preparedness level: take the average 7-day rolling POI visits during a window that begins after the first TC's landfall and ends before the second TC's preparedness period (e.g., the week starting one day after first landfall). If the landfall-sequence coefficient and the interaction term remain statistically significant with materially similar magnitudes, the concern is mitigated. Additionally, run a placebo test applying the same alternative baseline to a period with no second TC to quantify recovery-driven visit elevation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the second TC receives higher baseline preparedness rests on a definition of 'preparedness level' that may be contaminated by recovery activity. In Methods, the baseline is 'the average of the 7-day rolling average from four weeks to one week before the sequential TC events' — i.e., before the first storm. For the second TC, the preparedness window is the 7 days before its landfall, which can overlap with restocking, debris cleanup, and rebuilding after the first TC. If visits to gas stations, grocery stores, and building-material dealers remain elevated for reasons unrelated to the impending second storm, the ratio (peak 7-day rolling average / pre-first-storm baseline) overstates preparedness for the second TC. This directly inflates the landfall-sequence coefficient (Table 1, Model 2: 0.0202) and also biases the interaction term if post-storm activity correlates with the second TC's forecast wind. The paper does not test alternative baselines measured after the first storm, and the limitations section does not mention this baseline choice. Because the 'higher preparedness for the second TC' is a headline result, this is a load-bearing unaddressed risk.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper uses county-level mobility data from Dewey Inc. to measure situational preparedness for six pairs of sequential tropical cyclones that made landfall in the same U.S. state within 21 days during 2020-2024. Preparedness is operationalized through elevated visits to gasoline stations (mobility needs), grocery stores (daily supplies), and building-material dealers (structural reinforcement), relative to a baseline. The authors fit linear spatial Durbin models and spatial logistic models to show that: stronger forecast wind is associated with higher preparedness; the second TC receives higher baseline preparedness but has a flatter response to forecast wind; first-TC power outages and preparedness levels predict second-TC preparedness, with spatial spillovers; and counties with more children, elderly, limited English proficiency, and other access/functional-need populations show lower preparedness.","tokens_in":14526,"tokens_out":5791,"duration_ms":46931,"significance":"If the results hold, this is the first systematic empirical documentation of preparedness dynamics for sequential tropical cyclones, a growing hazard class. The use of high-frequency mobility data across multiple events is a strength, as are the complementary linear and logistic specifications and the spatial models that explicitly account for county interdependence. The findings on AFN disparities and infrastructure spillovers have clear policy relevance for emergency management. However, the headline 'second-TC preparedness is higher' rests on a baseline choice that may conflate recovery activity with forward-looking preparedness, and the inference is based on a modest number of event pairs; the paper would be substantially strengthened by robustness analyses that address these points.","major_comments":[{"comment":"The preparedness baseline is defined as the average of the 7-day rolling average from four weeks to one week before the sequential TC events—i.e., before the first storm—and is used for both TCs. For the second TC, the numerator is the peak 7-day rolling average in the seven days before its landfall. If visits to gas stations, grocery stores, or building-material dealers remain elevated after the first storm because of recovery, restocking, or repair activity, the second-TC ratio will overstate preparedness attributable to the second storm. This directly inflates the landfall-sequence coefficient (0.0202 in Table 1, Model 2) and can bias the interaction term if post-storm activity correlates with the second TC's forecast wind. The manuscript does not test an alternative baseline measured after the first storm (e.g., the week immediately preceding the second preparedness window) and does not list this choice among the limitations. Please add a robustness check with a post-first-storm baseline and discuss whether the qualitative conclusions survive.","section":"Methods, 'Mobility data' paragraph; Table 1, Model 2"},{"comment":"The regression data contain two observations per county (one for each TC in the pair), yet the models use state fixed effects and spatial lags without clustering standard errors by county or including county-level random effects. The two observations from the same county are likely correlated through shared demographics, infrastructure, and the common baseline; this can understate standard errors and inflate the significance of the headline coefficients. Please report cluster-robust standard errors (at the county or event level) or fit a panel specification, and indicate whether the conclusions in Tables 1 and 2 are robust.","section":"Statistical analysis; Tables 1 and 2"},{"comment":"The analysis of first-TC impacts on second-TC preparedness (Table 2) excludes the Helene-Milton pair because of missing 2024 power outage data. This is the most prominent sequential event in the study window and the motivating example of the Introduction. With only five event pairs (N=613), the power-outage and spillover estimates are based on limited cross-event variation. The paper should either provide a sensitivity analysis that includes Helene-Milton with an alternative outage measure, or explicitly discuss how the exclusion may affect the generalizability of these findings.","section":"Results, 'First landfall TCs influence the preparedness for the subsequent TCs'; Table 2"}],"minor_comments":[{"comment":"There are two apparent typos: in the abstract, 'approximately 13% pct' should be 'approximately 13%,' and in the final paragraph of the results, the third coefficient (1.418) is attributed to 'mobility needs preparedness' but should be 'structural reinforcement preparedness.'","section":"Abstract; Results, 'The county-level aggregated preparedness pattern'"},{"comment":"The significance legend reads '***p<0.001, ***p<0.01, *p<0.05'; the second entry should be '**p<0.01'.","section":"Table 2 note"},{"comment":"Causal language such as 'the first TC heightened people's risk perception' and 'power outage experiences... would increase preparedness' goes beyond what the regression design can establish. Please soften these statements to associational language or explicitly acknowledge confounding by recovery activity, resource availability, and other unmeasured factors.","section":"Discussion, 'A higher level of preparedness...'"},{"comment":"The description of the baseline window is confusing: 'from four weeks to one week before the sequential TC events' with the parenthetical '21-day before the preparedness period.' Please clarify whether the baseline is the same for both TCs and specify the exact calendar relationship.","section":"Methods, 'Mobility data'"}],"recommendation":"major_revision","confidential_remarks":"The main concern is the baseline definition for the second-TC preparedness measure; if the robustness check with a post-first-storm baseline does not change the qualitative results, I would be satisfied. The exclusion of Helene-Milton from the first-TC-impact models is also worth highlighting in revision. The paper is otherwise well-structured and the statistical framework is clearly described."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nWorth a look: this is the first paper I know of that tracks situational preparedness across sequential tropical cyclones using mobility data, and the phenomenon is real and growing. The authors assemble six paired events across seven states and test whether forecast intensity, landfall sequence, first-storm impacts, and county demographics move visits to gas stations, grocery stores, and building material dealers. The core findings—higher preparedness for the second storm, weaker sensitivity to forecast wind for the second storm, and power-outage spillover into neighboring counties—are new and plausible, and the analysis is mostly careful. Linear spatial Durbin models are backed by logistic robustness checks, and the coefficient patterns are consistent. Credit where due: the data assembly is substantial (mobility, NHC forecasts, HURDAT2, Stage IV rainfall, EAGLE-I outages, Census), and the paper doesn't oversell the daily-supplies and structural-reinforcement results when they don't hold.\n\nWhere it gets soft: the stress-test note is right and it's load-bearing. The baseline for 'preparedness level' is the 7-day rolling average from four weeks to one week before the sequential events—i.e., before the first storm. The second TC's preparedness is then peak visits in the seven days before its landfall divided by that pre-first-storm baseline. If gas stations, grocery stores, and building-material dealers stay busy during recovery and restocking, the ratio for the second storm is contaminated by post-disaster activity. That directly inflates the landfall-sequence coefficient (0.0202, Table 1) and could bias the interaction if post-storm activity correlates with the second storm's forecast wind. The authors don't test any post-first-storm baseline, and the limitations section doesn't mention it. That's a fixable robustness check, not a fatal flaw, but it needs to be done before I'd trust the headline.\n\nOther soft spots: Helene-Milton, the most salient pair, is excluded from the second-storm analysis due to missing outage data; no code or processed data is released (Dewey data is commercial, but the derived county-level measures could still be shared); and the prose occasionally slips into causal language for observational regressions. These are minor-to-moderate.\n\nBottom line: the central research question is well motivated, the empirical work is real, and the main results are probably in the right direction, but the baseline choice is an unaddressed threat to the main claim. Send it to peer review, but the referee should ask for an alternative-baseline robustness analysis and a clear statement of the limitation. I'd read a revised version carefully.","headline":"First systematic look at preparedness for back-to-back hurricanes, but the second-storm preparedness measure is built on a pre-first-storm baseline that likely inflates the headline sequence effect.","tokens_in":15156,"tokens_out":2134,"would_cite":false,"duration_ms":19285,"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":"Second hurricane: higher preparation, muted risk response.","keywords":["sequential tropical cyclones","situational preparedness","mobility data","point-of-interest visits","spatial Durbin model","power outage spillover","access and functional needs","landfall sequence"],"falsifier":"Recompute the second-storm preparedness ratio using a baseline taken from the period after the first storm's landfall (for example, the week between the two storms); if the 'higher preparedness for the second TC' coefficient shrinks to zero or reverses sign, the claimed effect is an artifact of post-disaster activity rather than forward-looking preparation.","tokens_in":14143,"feed_emoji":"🌀","tokens_out":5480,"duration_ms":45362,"temperature":0.7,"pith_summary":"This paper asks how people's hurricane preparedness changes when two tropical cyclones strike the same region within weeks. Using county-level mobility data showing visits to gas stations, grocery stores, and building-materials dealers before six sequential US landfalls (2020–2024), it argues that preparedness is a combined function of forecast wind speed and whether the storm is the first or second in the sequence. The central finding is that people prepare more, on average, for the second storm, but each additional unit of forecast wind moves them less than it did for the first storm. The paper also finds that a power outage during the first storm raises preparedness for the second, with effects spilling into neighboring counties. If these results hold, risk communication for back-to-back storms should expect elevated but partially fatigued audiences.","feed_headline":"Second hurricane: higher preparation, muted risk response","feed_subtitle":"Mobility data from six back-to-back US landfalls shows forecast wind moves people most before the first storm.","key_machinery":"The central machinery is a county-level measure of preparedness built from daily point-of-interest visit counts: a 7-day rolling average of visits to gasoline stations (mobility needs), grocery stores (daily supplies), and building-materials dealers (structural reinforcement), with a baseline defined as the average from four weeks to one week before the sequential event. Preparedness level is the ratio of the peak 7-day average in the week before landfall to that baseline; preparedness pattern is a ratio exceeding two standard deviations. These measures feed a spatial Durbin model that includes landfall sequence, forecast wind speed and their interaction, along with sociodemographic, access-and-functional-needs, infrastructure, and spatial-lag controls. The interaction term between forecast wind and sequence is what carries the claim that the second storm is prepared for more but heeded less.","core_discovery":"On the paper's own terms, the discovery is that situational preparedness for sequential tropical cyclones is governed by an interaction between objective hazard information and recent experience. In the mobility-needs measure (visits to gasoline stations), each additional meter per second of forecast wind raises preparedness by about 0.34%, the second landfalling TC receives about 2.02% more preparedness than the first, and the interaction term shows that the wind slope is about 0.12% per m/s shallower for the second storm. First-storm power outages raise second-storm preparedness by roughly 13% for mobility needs and 24% for structural reinforcement, and these effects spill over to neighboring counties. The paper interprets the higher baseline for the second storm as heightened risk perception after the first event, and the shallower wind slope as psychological fatigue, resource depletion, or anchoring on the first storm's experience.","pith_inferences":["Inference: the paper's 'second storm higher preparedness' estimate rests on a baseline set before the first storm; if recovery activity kept POI visits elevated after the first landfall, part of the 2.02% effect may be post-disaster activity rather than forward-looking preparation, and recomputing with a post-first-storm baseline would separate these.","Inference: the same POI-visit framework could be applied to other compound hazards—sequential floods, heatwaves, or wildfire followed by rain—to see whether the 'more preparation but duller response' pattern generalizes beyond tropical cyclones.","Inference: the interaction effect suggests a behavioral model where prior experience anchors risk perception; this could be tested against survey-based risk-perception data collected between two storms.","Inference: because the study uses visit counts rather than purchase amounts or dwell time, the true preparedness response may be understated for grocery and building-material stores; point-of-sale data would provide a higher-fidelity test."],"forward_implications":["Counties that prepare strongly for the first storm also prepare strongly for the second, suggesting preparedness is a stable county-level trait or habit.","Power outage experience is a stronger driver of subsequent preparedness than experienced wind speed, pointing to infrastructure disruption as the salient risk signal.","Spatial spillovers mean one county's power outage raises its neighbors' preparedness, so regional coordination can amplify or smooth preparedness.","AFN populations—children, elderly, and limited-English households—show persistently lower mobility-preparedness in sequential events, implying targeted outreach is needed.","Risk communication for the second storm must counter the diminished response to forecast intensity."],"supporting_citations":[{"why":"Establishes that sequential TC landfalls occur and are becoming more frequent, motivating the study.","marker":"[1]"},{"why":"Supplies the earlier finding that a prior hurricane affects preparedness for the next, and the POI-based method used here.","marker":"[13]"},{"why":"Shows POI visits rise before a hurricane, grounding the mobility-based preparedness measure.","marker":"[27]"},{"why":"Provides the forecast track and intensity archive used for forecast wind speed.","marker":"[29]"},{"why":"Provides the daily POI visit counts from which preparedness levels are computed.","marker":"[61]"},{"why":"Wind field model used to estimate county-level experienced wind from best-track data.","marker":"[63]"},{"why":"Power outage data used to quantify first-TC outage experience and its spillovers.","marker":"[69]"},{"why":"Census data for sociodemographic and access-and-functional-needs controls.","marker":"[70]"}],"fun_headline_variants":["Second hurricane: more prep, weaker wind response","Back-to-back hurricanes: higher prep, muted wind sensitivity","Sequential storms: outages boost prep and spill over","First storm outages prepare neighbors for second landfall","Wind moves first storm prep; second storm gets baseline boost"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The measure of second-storm preparedness uses a baseline from before the first storm, so any visits that remain elevated because of recovery, restocking, or repair after the first storm are counted as preparedness for the second rather than separated from it.","fun_headline_variants_meta":{"raw":{"variants":["Second hurricane: more prep, weaker wind response","Back-to-back hurricanes: higher prep, muted wind sensitivity","Sequential storms: outages boost prep and spill over","First storm outages prepare neighbors for second landfall","Wind moves first storm prep; second storm gets baseline boost"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000334,"raw_usage":{"total_tokens":1841,"prompt_tokens":921,"completion_tokens":920,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":537,"completion_tokens_details":{"reasoning_tokens":844}},"tokens_in":537,"tokens_out":920,"duration_ms":8953,"temperature":1.0,"reasoning_tokens":844,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:53:30.473981+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the second-storm preparedness ratio using a baseline taken from the period after the first storm's landfall (for example, the week between the two storms); if the 'higher preparedness for the second TC' coefficient shrinks to zero or reverses sign, the claimed effect is an artifact of post-disaster activity rather than forward-looking preparation.","supporting_citations":[{"cited_title":"& Lin, N","cited_arxiv_id":null,"evidence_quote":"Establishes that sequential TC landfalls occur and are becoming more frequent, motivating the study."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the earlier finding that a prior hurricane affects preparedness for the next, and the POI-based method used here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows POI visits rise before a hurricane, grounding the mobility-based preparedness measure."},{"cited_title":"S., Li, Q., Jawer, G., Xiao, X","cited_arxiv_id":null,"evidence_quote":"Provides the forecast track and intensity archive used for forecast wind speed."},{"cited_title":"& Renaud, F","cited_arxiv_id":null,"evidence_quote":"Provides the daily POI visit counts from which preparedness levels are computed."},{"cited_title":"& Smith, J","cited_arxiv_id":null,"evidence_quote":"Power outage data used to quantify first-TC outage experience and its spillovers."},{"cited_title":"https://hurricanes.ral.ucar.edu/repository/","cited_arxiv_id":null,"evidence_quote":"Census data for sociodemographic and access-and-functional-needs controls."}],"review_version":1}