Winter is Coming… for 90% of All Micromobility Markets

Zoba
Zoba
Dec 17, 2019 · 7 min read

This is our first guest post by Tarani Duncan, strategic product advisor at Zoba. Tarani is a micromobility industry veteran. Previously Motivate, JUMP, Mapbox, Tarani obsesses over day-to-day operations, enjoys spending time in the mountains, and can be found on Twitter at @taraniduncan.

Zoba provides demand forecasting and optimization tools to shared mobility companies, from micromobility to car shares and beyond.

This winter the budding micromobility industry will experience the most severe seasonal impact to business to date, which is why I partnered with Zoba to forecast the impact of seasonality on 50 operators across 248 markets globally. Below, I discuss methodology for creating the global seasonality scale. I also demonstrate how Zoba’s optimization toolkit helps operators develop strategies around right-sizing and optimally distributing their fleets during the frosty months ahead.

Forecasting Seasonal Changes in Demand

A day of surface temperatures as the earth rotates against the sun. Matthew Irwin, Mapbox.

From a forecasting perspective, seasonality refers to regular-sequenced, predictable changes that recur annually, like seasonal changes in weather patterns. For this particular seasonality analysis, I produce rough order-of-magnitude predictions of seasonal effects in 248 cities across the world.

Here’s the plan: Start with worldwide monthly weather datasets, including precipitation and temperature. Pick a seasonal operation with rich publicly available data and use this operator’s data to train a model which predicts relative changes in demand as a function of weather. Feed market weather data into the model to get rough estimates of how demand will change throughout the year. This approach hinges upon the strong assumption that people around the world react similarly to weather changes.

For weather data, I used 2018 global temperature data from Berkeley Earth, global precipitation data from NOAA. For general micromobility market data, like location and operator presence, I used a dataset compiled by grow.io. I converted NetCDF’s to geojsons and paired each market with historic data from the nearest weather stations to generate a list of monthly forecasted averages for precipitation, temperature, and snow through November of 2020. For the purpose of this analysis, I assumed weather patterns in 2018 to be consistent through 2020 (ie January 2018 has the same weather as January 2020).

For training data, I used ridership data from Citibike, New York City’s bike share program. Using this data, I modeled the relationship between precipitation, temperature, and ridership. I selected Citibike data to train my model for two reasons: the abundance of open source data — millions of trips segmented across membership type — day passes (tourists/recreation) and annual members (commuters) — and the clear seasonality of New York City.

Of course, there are limitations with the training dataset: most obviously, a bike is a different form factor than a kick scooter. With larger tires, a bike may be better suited for harsher weather. Citibike is also asset dense and dock-based, not loosely free-floating so usage patterns vary. It’s also worth noting the system’s hardware constraints: Citibike docks can get snowed in, resulting in occasional system-wide shutdowns. Additionally, New York City doesn’t have a monsoon season. So, the relative impact of precipitation may vary between New York and more tropical parts of the world, like Malaysia and Thailand.

While utilization is not demand, in the case of asset-dense Citibike, monthly system-wide utilization and demand are highly correlated exactly because the system has enough bikes to capture a high percentage of the demand. By modeling rides per vehicle as a function of monthly weather, we can estimate how demand in New York responds to weather variation. Under the strong assumption that people worldwide react somewhat similarly to changes in weather, I used the NYC-trained model to estimate relative seasonal changes in ridership worldwide. In spite of these limitations with the data, I believe we can use this model to create a general understanding of how seasonality impacts ridership patterns in active shared micromobility schemas around the world.

Predicted number of rides per vehicle per day against actuals

In order to create a market ridership model, which forecasts trips per vehicle per day, I used historic Citibike data to model rides per vehicle per day as a linear combination of temperature, precipitation, snowfall, and interaction terms thereof. The outputs of this model were the most statistically significant, capturing the major ups and downs in daily ridership as demonstrated in the chart above.

Developing the Seasonality Scale

We want an easy to understand way of forecasting the impact of seasonality. Feeding monthly weather data for each market through the model we trained on Citibike data yielded market-wise monthly predicted ridership. Each month’s score is (mi — mmax) / mi, where m is the predicted monthly ridership for the i-th month and mmax is the highest predicted monthly ridership across the 12-month period. Thus the monthly score represents what fraction of peak demand is lost due to weather.

Average global reduction in ridership per month

In the seasonality scale, 0 represents no change in demand due to seasonal weather changes whereas -1 represents a 100% decrease in demand for an operator’s service due to seasonal changes in precipitation, snow, and temperature. In the chart above, I average the monthly scores across 248 markets. Since most markets are in the northern hemisphere, the weather tends to be best May through August when average seasonality scores are closest to 0. Some of the markets included in this calculation have moderate climates year-round and thus display relatively little seasonality — think of places like Singapore and San Diego. Therefore, the cross-market average seasonality scale indicates gentler seasonal changes than many markets experience. Many northern European markets, for example, can expect to see January demand 80% lower than the summer peak. What did we find in our analysis? A staggering 60% of all global markets are likely to lose more than half their monthly revenue come February.

Deploying Winter Weather Tactics

In summary, this analysis of 50 operators across 248 markets globally found winter will impact 90% of all markets with at least a 20% decrease in monthly revenue. Zoba provides weather-tuned tooling operators can use to right-size and position their fleets in order to maximize rides, revenue, and accessibility during the most difficult times of year. Zoba’s winter weather toolkit includes dynamic fleet-sizing and real-time vehicle distribution recommendations.

  1. Dynamic Fleet Sizing: Matching Fleet Size with Seasonal Demand
    As this analysis suggests, shared micromobility is a highly seasonal business. In seasonal markets, operators see vehicle utilization dwindle while operations costs, like battery swaps, hold steady across the fleet. Operators also experience an increase in escalations per trip, meaning more inbound reports of mechanical issues across fewer trips taken. They also see unnecessary wear and tear on underutilized assets. In the winter, the cost of maintaining a larger fleet can be bad for business. Right-sizing a fleet means ensuring operators match fleet size with available demand to capture as much revenue as possible while preserving service quality and overall longevity of their assets.
  2. Optimal Distribution of Vehicles
    Commute patterns change seasonally. In the winter time, trips taken on shared mobility devices are necessary trips, like commuting from the bus to work. In the summertime, more recreational trip patterns emerge. This manifests locally in new seasonal demand patterns. Zoba’s optimal distribution tooling empowers local operators to place vehicles exactly where they’re needed in order to maximize rides, revenue, and accessibility.

Today marks a tipping point for shared mobility systems all over the world. In 60 days, the budding micromobility industry will experience the most severe seasonal impact to business to date. What strategies can operators deploy to maximize rides, revenue, and accessibility during the roughest months of the year? To what extent will specific companies experience the impact of seasonality and where will those companies feel the most operational pain? Zoba is currently helping operators make market-specific winter adjustments to fleets around the world.

Zoba is developing the next generation of spatial analytics in Boston. If you are interested in spatial data, urban tech, or mobility, reach out at zoba.com/careers.

Zoba Blog

Zoba increases the profitability of mobility operators through decision automation.

Zoba Blog

Zoba uses demand forecasting and optimization to improve the performance of shared mobility services. On this blog, Zoba operations leaders, data scientists, and engineers write about the problems we solve for shared mobility operators and tools we use to solve those problems.

Zoba

Written by

Zoba

Zoba increases the profitability of mobility operators through decision automation.

Zoba Blog

Zoba uses demand forecasting and optimization to improve the performance of shared mobility services. On this blog, Zoba operations leaders, data scientists, and engineers write about the problems we solve for shared mobility operators and tools we use to solve those problems.