Climate Risk in Mobility Apps: Measuring Is Not Mitigating
The rain starts and the app shows double, triple the fare, sometimes more. The user curses the app. The driver accepts the ride anyway, because he also needs the day to pay off. No one there can say, precisely, how much of that specific rain, in that specific corridor, explains the price that appeared on the screen.
Climate risk in urban mobility apps is the chance that rain, heat or flooding will reduce the supply of workers and increase wait time in a specific corridor, enough to trigger dynamic pricing, cancellation or missed deadlines. It is not uniform across the city: it is born and dies within a few blocks, and that is exactly where most platforms still fail to see it.
This applies to passenger transport and applies equally to food delivery apps and deliveries in general. They run on the same road network, at the same peak hours, under the same localized rain, with the same motorcyclist or cyclist as the most exposed asset.
- The impact of rain on mobility only became a measurable number after the app digitized ride times and fares: without historical series, it was perception, not measurement (Uber Newsroom, 2019).
- In Brazil, dynamic pricing entered the IBGE methodology in 2026, and app rides rose 56%, with spikes of up to 70% attributed to heavy rain (Exame, 2026).
- International literature diverges on the sign of the rain effect: in Chicago Uber demand rises 22%, in New York it falls. The cause is not the climate, it is the resolution of the measurement (ScienceDirect, 2025 and 2026).
- In urban basins, correlation between rain gauges drops precisely during heavy rain, exactly the rain that most affects operations (MDPI, 2020).
- In New York, 1.5 additional minutes of wait predicts a 37% drop in demand: the threshold that decides fleet reallocation can be measured in minutes, not millimeters (ScienceDirect, 2026).
- Bike and motorcycle delivery workers are among the groups most exposed to extreme heat, working outdoors with no cabin (Nexo, 2024; WRI Brasil, 2025).
In Campo Grande, two weeks ago, a storm made rides worth R$ 25 reach nearly R$ 60, and residents reported difficulty getting home according to a Campo Grande News report. Scenes like this repeat in Brazilian cities during summer rain. What separates a mature operation from another is knowing, hours in advance, that a given corridor would cross the critical threshold.
Operational climate risk, in my vocabulary, differs from physical climate risk. It is not the rain itself: it is the rain crossing a threshold that changes the behavior of supply, demand or price of a specific asset. In urban mobility, that asset is the corridor, the region, the time window. That is what this text is about.
Why the rain effect only became a number after operational data
The effect of rain on urban mobility could only be quantified after apps accumulated years of ride data, travel times and fares. Before that, the relationship between rain and traffic was shared intuition, not measurement. Operational data gave a ruler to environmental data, not the other way around.
Uber took three years of historical travel time data in São Paulo to state something precise: on days with rain above 0.1 mm per hour, traffic slowdown was on average 3.9 points higher than on days without rain, according to data published by the company itself. Before this series, no one in operations could say "how much," only "yes, it does."
This historical order made sense. Without a baseline for comparison, any statement about the rain effect would be a driver's opinion, not risk management. The app first needed to create its own thermometer of demand and supply to then have something to compare against the weather.
Today this effect is a visible part of the city's economy. Dynamic pricing entered IBGE's inflation calculation methodology in 2026, and app rides rose 56%, with spikes close to 70% in capitals with higher demand, according to an Exame report. 99 itself cites rain, alongside heavy traffic and demand peaks, as a trigger for higher fares.
The problem is that "digitizing demand" solved only half the equation. It says it rained and the price went up. It does not say on which corner driver supply disappeared first, nor how many extra minutes the passenger waited in that specific neighborhood before giving up. This gap is the subject of the next step.
Measuring is not mitigating: the two leaps, in time and in space
Measuring the rain effect on mobility is retrospective and aggregated: it explains what already happened, on average, across the entire city. Mitigating is prospective and local: it requires knowing in which corridor, in which time window, rain crosses the threshold that drops supply. These are two distinct leaps, and neither is solved with the same data.
The first leap is temporal: moving from "it rained yesterday and demand rose" to "it will rain at 5pm in this region and supply will fall." The second is spatial: moving from the city average to the specific corridor where the operation actually breaks. A platform can get the first leap right and still get the second wrong.
This is exactly what the case of Porto Alegre shows, where the same ride tripled in price during recent storms, according to a Diário Gaúcho/ClicRBS report (2026). The app knew it was raining in the city. It did not necessarily know, with minutes of advance notice, that that specific neighborhood would empty of drivers first.
This gap between measuring and mitigating has a measurable cost in other urban infrastructure sectors. In São Paulo, the City Hall maps 61 recurring flooding points, and each flooding point costs the city about R$ 1 million per day, with floods totaling R$ 762 million per year, according to a study by Haddad and Teixeira published by Agência FAPESP (2013).
For fleet operators, the parallel is direct. I have already written about how the cost of downtime caused by rain accumulates silently when no one measures the right corridor. In urban mobility, the "downtime" is not the physical fleet: it is the driver who pulls back, the passenger who cancels, the price that rises without anyone deciding on it.
Literature does not agree on the sign of the rain effect, and that is the clue
International studies disagree on whether rain increases or decreases demand for app rides. In Chicago it increases Uber demand by 22%. In New York, rainy days appear associated with lower demand. This divergence is not researcher error: it is a symptom that "it rained in the city" is too coarse a variable.
In the Chicago study, rain raised taxi rides by up to 5%, while Uber rose 22% and Lyft 19%, according to the study published in ScienceDirect. The authors attribute the difference to dynamic pricing: it incentivizes platform drivers to meet excess demand, something the taxi's fixed fare does not do.
In New York, on the other hand, another study published in ScienceDirect finds the opposite pattern: rainy days associated with lower demand. In Japan, models indicate that passengers tend to request rides precisely when it is not raining [FONTE_A_CONFIRMAR]. Three cities, three different answers to the same question.
The more honest explanation is not that one city "reacts differently" from another. It is that "it rained" aggregates intensities, durations and micro-regions with opposite responses within a single binary label. The result of each study depends on what dominated the sample, not on any universal law of human behavior under rain.
The authors of the Chicago study themselves acknowledge this limitation and recommend that future work treat rain intensity with more granularity, in addition to investigating supply shifts between areas of different income levels. This recommendation is, in practice, a request for finer resolution, exactly what is missing in most dispatch systems today.
The same Chicago study shows spatial heterogeneity within the city itself: the Loop and the areas immediately north and south register from 1.84 to 4.67 additional rides per thousand inhabitants during rain, versus a much smaller effect in the rest of the city. The city average disappears exactly where the effect is strongest.
Why fine resolution, and not just more data
The problem with measuring rain in urban mobility is not lack of data volume, it is the spatial scale of rain itself. Convective storms, the ones that most affect operations, occur within a radius of a few kilometers. A sparse sensor network tends to miss precisely the core of the event that matters most.
A study in an urban basin of 125 km² with 22 rain gauges showed that correlation between stations remains good for low and intermediate intensity rain, but drops significantly for heavy rain, according to research published in MDPI (2020). The maximum precipitation per area decays according to a power law between 1 and 30 km.
In practice, this means that two stations a few kilometers apart can register completely different rainfall during the same storm. The airport station may register light drizzle while, eight blocks away, a convective cell floods the avenue and paralyzes the corridor.
This is the final argument against the city average: the event with the greatest operational impact is precisely the one with the smallest spatial scale. Weak, homogeneous rain, well described by the average, is precisely the rain that does not break the operation. The localized downpour that the average erases is the one that drops supply.
Coarse resolution, then, is not just less precise in abstract terms. It systematically fails exactly where it matters most to get it right. I have detailed this point in another text about the difference between weather forecasting and climate risk management: a city-level forecast informs the climate, not the asset's risk.
i4sea works with a resolution of 1 to 3 km, versus about 25 km in traditional public forecasts, covering 18 hydrometeorological hazards with more than 10 years of historical reanalysis. This difference in scale is what separates "it will rain in São Paulo" from "it will rain heavily in this corridor, in this window."
The asset threshold: translating climate into minutes of waiting
What mobility platforms lack is not weather forecasting, that has already become a commodity available on any cell phone app. What is missing is impact forecasting: the translation of "it will rain X mm/h in this corridor" into "supply drops Y% and wait time doubles." The threshold belongs to the asset, not to the climate.
The right currency for this translation may not even be millimeters of rain. In New York, 1.5 additional minutes of delay already predicts a 37% drop in demand, according to the same ScienceDirect study (2026) cited above. Wait time weighs more on the passenger's decision than the ride price.
This changes the question the operation should ask. Instead of asking "how many millimeters will fall," the platform should ask "in which corridor does the forecast rain push wait time beyond the point where the passenger cancels." It is a behavioral threshold, not a purely meteorological one.
The cost of getting this translation wrong shows up in concrete numbers. A survey cited by Autopapo (2026) shows that 48.09% of cancellations in transport apps happen due to delay in arrival. If wait time is the variable that most drops demand, it is this variable that needs to be forecast in advance, corridor by corridor.
I have already written about this type of silent disruption in another operational context: unplanned downtime caused by extreme rain has the same shape in an urban mobility fleet and in an industrial asset. The difference is the time scale: in the app, the decision needs to be made in minutes, not hours.
Anticipating the event costs less than reacting to it. For a mobility platform, this means reallocating fleet before rain hits the critical corridor, not adjusting fares after the waiting queue has already formed. It is the difference between managing risk and simply passing the cost on to the passenger.
Food and deliveries: the same corridor, the most exposed worker
The reasoning does not apply only to passenger transport. Food and deliveries in general use exactly the same road network, at the same peak hours, with the same kind of localized rain. What changes is the currency of the loss: instead of the passenger's wait time, what breaks is the delivery deadline and the canceled order.
The supply and demand logic also repeats, with one detail that makes it worse. When rain intensifies, more people stay home and delivery demand rises. At the same time, some delivery workers stop for safety, because a flooded street is a direct risk for someone on a motorcycle or bike. Demand rises and supply falls, the same imbalance that passenger transport reveals through dynamic pricing and cancellation.
There is a difference that order volume does not show. In passenger transport, the asset is the car and the fleet. In delivery, the main asset is the person. Bike or motorcycle delivery workers work outdoors, with no cabin and no roof, and appear among the groups most exposed to extreme heat, according to a survey by Nexo with WRI Brasil. Dispatching a delivery to a flooded stretch stops being just a pricing decision and becomes an operational and worker safety decision.
That is why the delivery operator's question is the same as the fleet operator's: in which corridor, in which window, does the forecast rain cross the threshold that drops supply and blows the deadline. The scale that answers this question is the same, and it is by corridor, not by city. When the corridor becomes the unit of decision, the delivery platform stops discovering the problem through the order that was late and starts acting before the peak.
Free climate risk diagnosis: see, corridor by corridor, where the city-level forecast hides the risk your operation already feels in wait time and cancellation.
Frequently asked questions
Why does a ride become so much more expensive when it rains?
Because driver supply drops exactly when demand rises, and the dynamic pricing algorithm reacts to this imbalance. In Brazil, this mechanism has raised rides by up to 56%, with peaks of 70% in capitals, according to Exame (2026).
Isn't rain forecasting enough to avoid the problem?
Not alone. Rain forecasting says it will rain in the city; it does not say in which corridor supply will drop below the critical point. What is missing is the translation of millimeters of rain into minutes of waiting, which is what decides fleet reallocation.
What is operational climate risk in mobility?
It is the chance that a climate condition, such as heavy rain in a specific corridor, crosses the threshold that reduces driver supply and increases wait time enough to trigger dynamic pricing or cancellation. It is not the climate itself, it is the climate crossing a business threshold.
Why doesn't a city-level forecast solve the corridor problem?
Because convective storms occur at scales of a few kilometers, and correlation between measuring stations drops precisely during heavy rain, according to a study published in MDPI (2020). The city average erases exactly the event that matters most to the operation.
Do food and delivery apps suffer the same rain effect?
Yes, and the logic is the same. They use the same road network and face the same type of localized event, with delivery worker supply falling while delivery demand rises. What changes is the currency of the impact: blown deadlines and canceled orders instead of passenger wait time, with the delivery worker as the most exposed asset (Nexo, 2024; WRI Brasil, 2025).
What changes when the corridor becomes the unit of decision
As long as rain is treated as a city-wide fact, the platform will keep reacting with dynamic pricing after the passenger has already given up. When the corridor becomes the unit of measurement, the decision shifts from "how much to charge" to "where to reallocate fleet before the threshold is crossed." In a delivery operation, the same shift changes the question from "how much to charge for delivery" to "where to dispatch before the peak."
If you operate fleet, dispatch, demand planning or a delivery operation, I invite you to take i4sea's free climate risk diagnosis, point by point, to see where the city-level forecast hides the real risk of your corridor. Request your diagnosis here.
References
Uber Newsroom (2019). Traffic and rain data in São Paulo. https://www.uber.com/pt-BR/newsroom/uber-compartilha-dados-para-ajudar-pesquisas-sobre-transito-e-em-politicas-de-mobilidade
Exame (2026). Por que corridas por aplicativo ficaram 56% mais caras. https://exame.com/economia/por-que-corridas-por-aplicativo-como-uber-e-99-ficaram-56-mais-caras/
Campo Grande News (2026). Corridas por aplicativo mais que dobram de preço após temporal. https://www.campograndenews.com.br/cidades/capital/corridas-por-aplicativo-mais-que-dobram-de-preco-apos-temporal
ScienceDirect (2025). Case study sobre chuva e demanda de corridas em Chicago. https://www.sciencedirect.com/science/article/pii/S2590198225004919
ScienceDirect (2026). Estudo sobre chuva e demanda de corridas em Nova York. https://www.sciencedirect.com/science/article/abs/pii/S0739885926000351
MDPI (2020). Spatial Rainfall Variability in Urban Environments. https://www.mdpi.com/2073-4441/12/4/1157
Diário Gaúcho/ClicRBS (2026). Chuva triplica preço de corrida em Porto Alegre.
Nexo (2024). Entregadores por aplicativo e exposição ao calor extremo.
WRI Brasil (2025). Exposição climática de entregadores de bike e moto.
Autopapo (2026). Pesquisa sobre cancelamentos em apps de transporte.
Prefeitura de São Paulo. Mapeamento de pontos de alagamento recorrente.
Haddad, E. e Teixeira, E. Agência FAPESP (2013). Custo econômico das enchentes em São Paulo.
CNseg/EY, COP30 (2025). Perdas climáticas no Brasil e cobertura de seguro.
Ecodebate (2026). Impacto do calor na produtividade do trabalho.