Hurricane recovery gaps lie in reconnection, not local restart
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: We're looking at the paper: "Hurricane recovery gaps lie in reconnection, not local restart".
Mira: Here's the gist of the paper. Hurricane recovery gaps lie in reconnection, not local restart Analyzing device-based mobility for Hurricanes Helene and Milton, we find local activity restarted within 4.5 days in both storms, whereas reconnection to the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia The gap lies in external reconnection, not local restart, implying planners should target rural isolation Study Focus and Methodology This research analyzes human mobility data from Hurricanes Helene and Milton to determine if recovery gaps exist across the rural–urban gradient The study decomposes mobility into within-area, inflow, and outflow components across six activity categories and two spatial scales The analysis involved analyzing counties within 100 miles of each storm’s IBTrACS track for Helene and 487 counties for Milton, encompassing 42.9 million residents Mobility data were derived from device-based foot-traffic records aggregated into six activity groups: Travel, Work & Professional, Health, Education, Retail & Leisure, and Urban Government A counterfactual baseline was established using a calendar regression model to predict daily visit counts Disruption was quantified as the relative deviation from this counterfactual baseline, measured by the metric rd(t) = 100((z (t)) - 1) Recovery Dynamics Comparison The study compared recovery times between the two storms, finding that while within-area mobility recovered at similar rates, inflow recovery differed significantly Within-area mobility recovered in 3.2–5.1 days under Helene and 4.7–6.1 days under Milton However, incoming visits took substantially longer to recover after Helene, taking an average of 11.3–13.3 days compared to 4.4–5.5 days after Milton This delay in inflow recovery extended beyond Travel to visits associated with work, healthcare, and education The finding suggests that local activity can rebound while incoming travel remains depressed, leaving an incomplete picture of recovery Spatial and Socioeconomic Influences The analysis examined factors influencing mobility disruption, including storm exposure and local characteristics Under Helene, coastal areas experienced deeper within-area declines, heavier rainfall was associated with larger inflow declines, and a higher proportion of White residents was also associated with deeper inflow declines The percentage-White association weakened to near zero after accounting for broad spatial patterns in Helene’s inflow disruption, while the coastal association became stronger Pooled sensitivity analysis showed that a one-standard-deviation increase in income was associated with a 5.28-percentage-point deeper decline in inflow movement under Helene Implications for Planning The results reframe resource allocation by showing that initial mobility declines alone are insufficient to predict prolonged disruption Places with the largest initial mobility declines were not necessarily those with the longest recovery Recovery planning should track how quickly incoming travel returns to identify communities where local movement has resumed but connections with the wider region remain disrupted Persistent inflow deficits can inform where recovery support remains needed by flagging communities for checks of transport links and essential-service access The study concludes that distinguishing the resumption of local activity from the return of incoming movement provides a more complete measure of post-hurricane recovery Limitations Three limitations qualify these findings, including the fact that Helene and Milton affected regions had different storm conditions, geography, and settlement patterns Device-based visits do not identify trip purpose and may under-represent rural residents, older adults, and people with low incomes Coverage gaps and measurement disruption may affect recovery estimates in an uncertain direction Finally, the recovery measure estimates when visit counts return to expected levels; it does not establish whether that return persists or whether essential services have fully recovered Supplementary Data Details The study utilized a flow-specific threshold of 20,000 visits per day to retain fewer Helene clusters and Milton counties with different flow types Recovery estimates were calculated using a centered three-day moving average to locate the trough within 10 days after landfall and fitting a Theil–Sen line to its nondecreasing recovery segment Spatial adjustment was assessed by fitting generalized additive models with a penalized two-dimensional geographic smooth to test sensitivity to broad geographic patterns The pooled sensitivity analysis compared models without spatial adjustment with those containing separate spatial smooths for each hurricane The study also examined residual Moran’s I under Queen contiguity to assess residual dependence in Milton inflow, which showed a non-significant result The comparison of covariate coefficients between models without and with spatial adjustment demonstrates how associations change after allowing for broad spatial variation Key Findings Summary The canonical 10-day trough search yielded 89 valid Helene inflow estimates before volume restriction, compared to 72 in the earlier 7-day implementation The median inflow contrast is 6.0 days, or approximately 2.2 times as long under Helene, calculated from the rounded medians Milton has the deeper within-area and inflow deficits, while the outflow maxima have similar medians and substantially different interquartile spans The inflow recovery gap is therefore not a general ordering of recovery by storm or by disruption depth The cross-storm recovery contrast does not demonstrate a within-region rural–urban gradient or identify a causal effect of rurality The spatial smooth for Helene inflow has P = 0.023 and deviance explained increases from 0.15 to 0.52 The spatial adjustment increased deviance explained from 44.6% to 63.5% for inflow decline The estimates for White population share (β = 3.74, −2.23 to 9.72) and rainfall (β = −3.54, −7.13 to 0.04) remained uncertain The cross-storm recovery contrast does not demonstrate a within-region rural–urban gradient or identify a causal effect of rurality Table Summaries Table S3 shows that the median inflow contrast is 6.0 days, or approximately 2.2 times as long under Helene, calculated from the rounded medians The within-area recovery (days) showed no statistically significant difference between storms (Mann–Whitney P = 0.294) The inflow peak deviation (%) showed a statistically significant difference between storms (P < 0.001) Table S4 shows that the median inflow contrast is 6.0 days, or approximately 2.2 times as long under Helene, calculated from the rounded medians The mean within-area recovery times were similar between storms (4.7 days for Helene and 4.8 days for Milton) The median inflow contrast is 6.0 days, or approximately 2.2 times as long under Helene, calculated from the rounded medians The within-area contrast was −0.
Title and authors: Kai: So we’re diving into "Hurricane recovery gaps lie in reconnection, not local restart" by Liu and Cao. We already touched on the core idea that the paper looks at mobility after Helene and Milton to find where recovery is lagging. Now let's talk about what those authors are actually trying to say with that title.
Mira: They are arguing against a simpler idea—that recovery is just about getting people moving around locally within their immediate vicinity. They’re saying that this local movement can often resume even when the bigger picture, the connection back to the outside world, is still severely broken.
Lev: So what does that mean for disaster response? If local activity recovers fast, why do we still see long delays in getting things like healthcare or work back up? Is it just that people don't know how to get back to those external services?
Kai: Exactly. The paper suggests the gap isn't in the local movement itself—the within-area recovery is pretty quick under both storms—but rather in that external reconnection. It implies we should be targeting rural isolation for support because that’s where the biggest delay is happening.
Mira: So, they are re-framing the equity question. It’s not just about who can leave a disaster zone, but who is left waiting for those connections to open up again after local movement has settled down.
Lev: From an engineering standpoint, that means if we were designing recovery infrastructure, we wouldn't just build stronger local roads; we’d be looking at how to maintain or restore the links to regional transport networks.
Kai: It’s about distinguishing between a community that can function on its own and one that is still waiting for external input. That distinction matters for resource allocation.
Mira: And they did this by decomposing mobility into within-area, inflow, and outflow components across six different activity categories and two spatial scales to see where exactly these delays occur.
Lev: Decomposing it like that sounds like a necessary step, but what’s the biggest challenge in modeling that kind of movement? Capturing the actual physics of human decision-making under stress.
Kai: That’s true. And they did use device-based foot traffic records to get that data, which is a great source for measuring what people are actually doing every day, but you have to remember the limitations we talked about before—it doesn't tell you the purpose of every trip.
Mira: Right. So it’s a useful tool for seeing patterns, but it’s not a perfect replacement for direct surveys or other methods that capture intent. That uncertainty in measurement is something they flagged themselves as a limitation.
Lev: It sounds like the research provides a clear direction for what to measure next—focusing on that inflow recovery time specifically, rather than just the initial decline.
Kai: That’s the point, right? The paper gives planners a specific lens through which to view post-hurricane recovery: look at reconnection gaps instead of just local restart speeds. This sets up the next part where we look at their actual findings on those numbers.
The paper's summary: Mira: So now let’s get into the specifics of what the paper actually found regarding these two storms, Helene and Milton. It compares how quickly local movement recovered against how long it took for incoming visits to return after each event.
Kai: The summary shows that while within-area mobility recovered at similar rates between the two storms, inflow recovery took much longer after Helene compared to Milton. They found that within-area mobility recovered in about three point two to five point one days under Helene and four point seven to six point one days under Milton.
Lev: That sounds like a relatively quick turnaround for local movement, which is good news for immediate community stability, even though the inflow recovery time is what’s concerning us.
Kai: That’s right. But the inflow recovery was substantially slower after Helene. For incoming visits across all six categories they looked at, it took between eleven point three and thirteen point three days to return to baseline after Helene.
Mira: When you compare that to Milton, where inflow recovery took only about four point four to five point five days, that difference is significant—it’s almost twice as long after Helene.
Lev: That’s a big quantitative difference. So the paper is showing that local activity can rebound while incoming travel remains depressed for a much longer period, which leaves an incomplete picture of full recovery.
Kai: It really highlights that just looking at local movement isn't enough to judge the overall health of a community after a major event. The delay in inflow recovery extends beyond just travel; it hits work, healthcare, and education too.
Mira: For example, incoming health visits took about twelve point one days to recover after Helene, compared to about four point six days after Milton. That means even essential services are lagging significantly in their return to normal levels.
Lev: So the finding is that local movement recovers sooner than external connections, which is a key distinction for assessing recovery progress on the ground.
Kai: It confirms what the authors suggest: local activity can rebound while incoming travel remains slow, and this gap gives us a better measure of post-hurricane recovery. This whole thing is called "Hurricane recovery gaps lie in reconnection, not local restart."
Mira: So to put it simply, they are saying that the problem isn't people being stuck inside; the problem is people being cut off from what’s outside. That’s a powerful way to frame it for planning discussions.
The paper's improvements: Kai: Now let's look at how the paper itself suggests we should improve recovery planning based on these findings. It moves beyond just reporting the data and tells us what action to take next.
Mira: They suggest that resource allocation shouldn't just be based on which area showed the biggest initial drop in mobility, because those areas weren't necessarily the ones that took the longest to recover overall.
Lev: So they’re saying we need a different way to look at disruption—we should track how quickly incoming travel returns to identify communities where local movement has resumed but connections with the wider region remain disrupted.
Kai: That’s right. The improvement is shifting the planning focus from tracking the initial decline to tracking how fast external links come back online. This helps pinpoint exactly which communities still need help reconnecting with the larger system.
Mira: It's a refinement of risk assessment, too, because they suggest we should look at storm effects on different spatial scales and activities separately. They want emergency managers to differentiate between local recovery and regional connectivity issues during their assessment phase.
Lev: So they’re advising a layered approach: assess the local situation first to see if movement is resuming, then check the external links to see if that reconnection is happening. That makes sense for operational planning.
Kai: They also suggest using factors like storm exposure and local characteristics—like rainfall intensity—to help adjust expected mobility shocks for different areas. This allows models to account for those specific local exposures during prediction.
Mira: And they did a bit of sensitivity analysis on how things change depending on geography, showing that the way we see these patterns can change depending on whether we include broad spatial variation in our models or not.
Lev: So, they are emphasizing that the model assumptions themselves matter—how you smooth out those geographic patterns directly impacts what you conclude about the associations between things like income or rainfall and mobility disruption.
Kai: It sounds like the overall suggestion is to stop treating initial mobility declines as a simple predictor of long-term damage. Instead, we should use that data to identify where local activity has resumed but regional connections are still broken.
Conclusion: Mira: So wrapping up this discussion on "Hurricane recovery gaps lie in reconnection, not local restart" by Liu and Cao. They’ve shown that the gap between what happens locally and what happens externally is primarily about external reconnection delays.
Kai: The paper highlights that within-area mobility recovers relatively fast, but inflow recovery can be significantly slower after events like Helene or Milton, taking up to thirteen days in some cases compared to just a few days for local movement.
Lev: So the main implication for us is that when we assess a disaster area, we need to look beyond just local movement and actively check the health of those external connections.
Kai: That’s it. We need to track where those inflow deficits persist because that flags communities needing help with transport links and essential service access while local movement is already picking up again.
Mira: So they concluded that distinguishing the resumption of local activity from the return of incoming movement provides a more complete picture for understanding recovery dynamics after a major hurricane event. That’s the main message to carry forward.
Lev: I think for my work, this means focusing on building systems that support those long-distance connections, not just optimizing local flow within a confined space.
Kai: A solid way to end it. It gives us a clear focus for what to look at in the next piece of research and planning related to post-disaster recovery.
Scarsdale High School · Bartlett School of Environment, Energy and Resources, University College London · Environmental Health Sciences, Columbia University Mailman School of Public Health
physics.soc-ph, stat.AP
Submitted: 2026-10-13
Updated: 2026-11-05
Code: https://github.com/Qing1011/milton_helene_mobility_category_analysis
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 83/100
The gist: Hurricane recovery gaps lie in reconnection, not local restart Analyzing device-based mobility for Hurricanes Helene and Milton…
Key concepts
- Within-area Mobility
- This measures how people move around inside a specific geographic area affected by the storm. The study found that this type of movement tends to recover relatively quickly after a storm, often within a few days, indicating local life is resuming.
- Inflow Recovery
- This refers to the time it takes for people coming into an affected area from outside to return to normal travel patterns. The research showed this process is significantly slower than local recovery, highlighting a major gap in post-disaster recovery.
- Rural Isolation Targeting
- This is a planning strategy suggested by the findings. Since external reconnection is the main bottleneck, planners should specifically target rural communities where connections to the wider region are failing, focusing support on transport and essential services access.
Terminology
Summary
Hurricane recovery gaps lie in reconnection, not local restart Analyzing device-based mobility for Hurricanes Helene and Milton, we find local activity restarted within ~4.5 days in both storms, whereas reconnection to the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia The gap lies in external reconnection, not local restart, implying planners should target rural isolation
Study Focus and Methodology
This research analyzes human mobility data from Hurricanes Helene and Milton to determine if recovery gaps exist across the rural–urban gradient The study decomposes mobility into within-area, inflow, and outflow components across six activity categories and two spatial scales The analysis involved analyzing counties within 100 miles of each storm’s IBTrACS track for Helene and 487 counties for Milton, encompassing 42.9 million residents Mobility data were derived from device-based foot-traffic records aggregated into six activity groups: Travel, Work & Professional, Health, Education, Retail & Leisure, and Urban Government A counterfactual baseline was established using a calendar regression model to predict daily visit counts Disruption was quantified as the relative deviation from this counterfactual baseline, measured by the metric rd(t) = 100((z (t)) - 1)
Recovery Dynamics Comparison
The study compared recovery times between the two storms, finding that while within-area mobility recovered at similar rates, inflow recovery differed significantly Within-area mobility recovered in 3.2–5.1 days under Helene and 4. p. 0 Analyzing device-based mobility for Hurricanes Helene and
Milton, we find local activity restarted within ~4. 5 days in both storms, whereas reconnection to
the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. Places
with the largest initial mobility declines were not necessarily those with the longest recovery. Although movement within local areas recovered at similar rates after both storms, incoming
travel took substantially longer to recover after Helene. The 2024
Atlantic season offers an unusual natural comparison: Hurricanes Helene (landfall 26 September
2024) and Milton (9 October 2024) struck the southeastern United States 13 days apart, one
inland and rural, the other coastal and urban. Prior work compared their aggregate regional
mobility during the landfall week13 but did not measure recovery, resolve activity categories, or
examine local spatial structure.Where the paper says it
Where the paper says it
Where the paper says it
Spatial and Socioeconomic Influences
The analysis examined factors influencing mobility disruption, including storm exposure and local characteristics Under Helene, coastal areas experienced deeper within-area declines, heavier rainfall was associated with larger inflow declines, and a higher proportion of White residents was also associated with deeper inflow declines The percentage-White association weakened to near zero after accounting for broad spatial patterns in Helene’s inflow disruption, while the coastal association became stronger Pooled sensitivity analysis showed that a one-standard-deviation increase in income was associated with a 5.28-percentage-point deeper decline in inflow movement under Helene
Implications for Planning
The results reframe resource allocation by showing that initial mobility declines alone are insufficient to predict prolonged disruption Places with the largest initial mobility declines were not necessarily those with the longest recovery Recovery planning should track how quickly incoming travel returns to identify communities where local movement has resumed but connections with the wider region remain disrupted Persistent inflow deficits can inform where recovery support remains needed by flagging communities for checks of transport links and essential-service access The study concludes that distinguishing the resumption of local activity from the return of incoming movement provides a more complete measure of post-hurricane recovery
Limitations
Three limitations qualify these findings, including the fact that Helene and Milton affected regions had different storm conditions, geography, and settlement patterns Device-based visits do not identify trip purpose and may under-represent rural residents, older adults, and people with low incomes Coverage gaps and measurement disruption may affect recovery estimates in an uncertain direction Finally, the recovery measure estimates when visit counts return to expected levels; it does not establish whether that return persists or whether essential services have fully recovered
Supplementary Data Details
The study utilized a flow-specific threshold of 20,000 visits per day to retain fewer Helene clusters and Milton counties with different flow types Recovery estimates were calculated using a centered three-day moving average to locate the trough within 10 days after landfall and fitting a Theil–Sen line to its nondecreasing recovery segment Spatial adjustment was assessed by fitting generalized additive models with a penalized two-dimensional geographic smooth to test sensitivity to broad geographic patterns The pooled sensitivity analysis compared models without spatial adjustment with those containing separate spatial smooths for each hurricane The study also examined residual Moran’s I under Queen contiguity to assess residual dependence in Milton inflow, which showed a non-significant result The comparison of covariate coefficients between models without and with spatial adjustment demonstrates how associations change after allowing for broad spatial variation
Key Findings Summary
The canonical 10-day trough search yielded 89 valid Helene inflow estimates before volume restriction, compared to 72 in the earlier 7-day implementation The median inflow contrast is 6.0 days, or approximately 2. p. 0 Analyzing device-based mobility for Hurricanes Helene and
Milton, we find local activity restarted within ~4. 5 days in both storms, whereas reconnection to
the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. Analyzing device-based mobility for Hurricanes Helene and
Milton, we find local activity restarted within ~4. 5 days in both storms, whereas reconnection to
the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. Analyzing device-based mobility for Hurricanes Helene and
Milton, we find local activity restarted within ~4. 5 days in both storms, whereas reconnection to
the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. Analyzing device-based mobility for Hurricanes Helene and
Milton, we find local activity restarted within ~4. 5 days in both storms, whereas reconnection to
the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. Here we decompose mobility into within-area, inflow and
outflow components across six activity categories and two spatial scales and show that the
rural–urban recovery gap lies almost entirely in external reconnection rather than in local restart. We analyzed counties within 100 miles of each storm’s IBTrACS track14
, 487 counties for
Helene and 34 for Milton, together encompassing 42. The percentage-White association, however, weakened to near zero after accounting for broad
spatial patterns in Helene’s inflow disruption, while the coastal association became stronger (Fig. Adding geographic adjustment increased deviance explained from 15% to 52%. The percentage-White association, however, weakened to near zero after accounting for broad
spatial patterns in Helene’s inflow disruption, while the coastal association became stronger (Fig. Adding geographic adjustment increased deviance explained from 15% to 52%. Milton’s estimates were less conclusive: all intervals included zero, with uncertainty
remaining substantial in its smaller sample. The percentage-White association, however, weakened to near zero after accounting for broad
spatial patterns in Helene’s inflow disruption, while the coastal association became stronger (Fig. We fitted separate models for Helene and
Milton, allowing associations between mobility disruption, storm exposure and local
characteristics to differ between events (Fig. Under Helene, coastal areas experienced deeper
within-area declines, heavier rainfall was associated with larger inflow declines, and outflow
increases were smaller farther from the storm track.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Table Summaries
Table S3 shows that the median inflow contrast is 6.0 days, or approximately 2.2 times as long under Helene, calculated from the rounded medians The within-area recovery (days) showed no statistically significant difference between storms (Mann–Whitney P = 0. p. 3 4; n = 33), with no statistically significant difference between storms
(Mann–Whitney P = 0. In contrast, inflow recovery took more than twice as long after
Helene (median 11. 4; n = 33), with no statistically significant difference between storms
(Mann–Whitney P = 0. In contrast, inflow recovery took more than twice as long after
Helene (median 11. 4; n = 33), with no statistically significant difference between storms
(Mann–Whitney P = 0. In contrast, inflow recovery took more than twice as long after
Helene (median 11. 1 shows daily
deviation from the counterfactual baseline for each of six categories, from one week before
landfall to two weeks after. Within-area mobility fell by 24–45% across categories under Milton
against 8–13% under Helene, between two and four times deeper in every category, and inflow
fell by 27–60% against 20–27%. Analyzing device-based mobility for Hurricanes Helene and
Milton, we find local activity restarted within ~4. 5 days in both storms, whereas reconnection to
the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. In contrast, inflow recovery took more than twice as long after
Helene (median 11. 1 days, IQR 8. 1 shows daily
deviation from the counterfactual baseline for each of six categories, from one week before
landfall to two weeks after. Within-area mobility fell by 24–45% across categories under Milton
against 8–13% under Helene, between two and four times deeper in every category, and inflow
fell by 27–60% against 20–27%.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
improve disaster recovery planning by targeting rural isolation rather than focusing solely on local restarts, as
The gap lies in external reconnection, not local restart.
This allows planners to identify communities needing support for transport links and essential service access when local movement has resumed but connections with the wider region remain disrupted. -
enable more nuanced resource allocation by incorporating inflow recovery time alongside within-area recovery time, as
Places with the largest initial mobility declines were not necessarily those with the longest recovery.
This helps identify communities facing prolonged disruption where incoming travel remains depressed even after local movement has resumed. -
enhance risk assessment by distinguishing between storm effects on different spatial scales and activities, such as analyzing
within-area mobility recovered at similar rates
while noting thatinflow recovery took more than twice as long after Helene (median 11.1 days, IQR 8.7–13.4; n = 46 clusters) as after Milton (median 5.1 days, IQR 4.2–6.2; n = 28 counties).
This allows emergency managers to differentiate between local recovery and regional connectivity issues during post-hurricane assessment phases. -
improve predictive modeling of disruption by assessing the sensitivity of covariate associations to geographic patterns, as
Comparing the covariate coefficients between M0 and M1 assesses how the associations change after allowing for broad spatial variation in mobility disruption.
This helps researchers avoid misattributing observed disruptions solely to socioeconomic conditions when those conditions are confounded by geography. -
refine risk factors for specific activities by identifying storm-related drivers of disruption, as
Under Helene, coastal areas experienced deeper within-area declines, heavier rainfall was associated with larger inflow declines.
This allows for activity-specific risk profiling where models can adjust expected mobility shocks based on local exposure and environmental conditions.
Abstract
Climate-intensified hurricanes disrupt urban mobility, yet whether recovery is equitable across the rural-urban gradient is unclear. Analyzing device-based mobility for Hurricanes Helene and Milton, we find local activity restarted within approximately 4.5 days in both storms, whereas reconnection to the outside world took 5 days in coastal-urban Florida but 11 in inland-rural Appalachia. The gap lies in external reconnection, not local restart, implying planners should target rural isolation.
Related papers
- Transient Signal Propagation in Heterogeneous Networks under Various Inputs
- Phase Transitions in Economic Inequality:Taxation and Extremal Replacement Dynamics
- New rules, old habits: across seven team sports, each league keeps its own way of using the rules
- URBAN-SPIN: A street-level bikeability index to inform design implementations in historical city centres
- Logarithmic dispersion at marginal stability in collective attention
- Glassy dynamics of metropolitan human mobility frozen near the gravitational equilibrium