What if the best way to map how a country travels is to combine the reach of mobile networks with the detail of travel surveys? This paper builds that hybrid picture for every small area of England and Wales. Origin–destination, or OD, matrices record how many trips run between each pair of places.

They support transport model calibration, accessibility and catchment measurement, spatial interaction modelling, and the business case for new transport or housing. That work needs OD evidence at small-area resolution because the decisions themselves are local.

This expectation is now written into policy. Policy asks authorities to make place-based assessments of how people move within and beyond their administrative area, because travel-to-work areas frequently do not align with authority boundaries.

The evidence also needs to be segmented by mode and purpose: an intervention aimed at the school run is not the same as one aimed at peak commuting. Analysts rarely get both at once, and instead choose among sources that are strong in different places, keeping survey-based and passively-collected mobility research largely separate.

Census journey-to-work flows are open and spatially detailed, but they cover a single trip purpose and are collected once a decade. The National Travel Survey records mode, purpose and time of day in depth, but its 2024 round had 8,975 participating households, all in England.

Spread across the 7,264 MSOAs of England and Wales, that is roughly one household per zone, so the survey cannot support small-area OD estimation. Passive mobile network data observe movement at population scale and can support OD and travel-demand estimation, but they count device-carrying adults, infer mode from movement traces, and do not record trip purpose.

The mobile products are also licensed for defined uses, so a product derived from them cannot simply be passed on. The fused dataset integrates mobile and survey sources to produce dense MSOA-to-MSOA OD matrices for England and Wales, split into seven travel modes and representative time periods, with eight trip purposes for the weekday morning peak.

The mobile records contribute the observed spatial structure of movement at a resolution no survey sample can reach. The NTS, census populations and trip-production rates add travel by children, a survey-referenced mode mix, and trip purpose. The fusion runs as a staged calibration, with every stage implemented in the published pipeline so the evidence and assumptions behind any value can be followed in the code.

The pipeline is published so the matrices can be inspected, reproduced and rerun for other years or study areas. Five input streams feed the build. Only the BT mobile network aggregate is commercially licensed; the other four are open or available to researchers under standard licences.

Because only the mobile base is licensed, the pipeline can be rerun outside the original project by anyone holding a licence for a compatible feed. Reading across the three columns, each dataset’s content is paired with a specific pipeline function.

The table shows a progression from observed origin–destination patterns and geography standardisation, through regional calibration and population weighting, to purpose allocation and constraints on origin–period mode totals. The pipeline runs four sequential steps.

Each adds a dimension the mobile aggregates do not carry, or references one to survey evidence, and each step is switchable so sensitivity builds differ only in configuration. The trip volume records movement from origin MSOA to destination MSOA by mode and time period, while superscripts mark processing stages: as delivered, after the all-age uplift, and after mode calibration.

Calibration operates on groups of records rather than single records, so each record also carries its origin region and trip-length band. Because the BT mobile data are scoped to adults, volumes at each origin are uplifted to all ages using the local population ratio of adult and child residents.

The mean uplift in the national run is 1.226, corresponding to a mean origin child share of 18.3 percent. Added child travel inherits the adult OD pattern of its origin, a limitation the paper returns to later. Mode calibration is the core step where the dataset’s mode mix is set by the survey rather than by the mobile source.

The survey reports average trips by mode for each region of residence and trip-length band. The thirteen survey modes are collapsed into the four broad classes distinguished by the BT feed: walking, road, rail and subway. This creates a target mode share for each of nine English regions and eight trip-length bands.

Because the NTS has covered England only since 2013, the table provides no Welsh targets. The original BT data classify all motor-based modes as one broad class, ROAD. That class is split into cycle, private car, motorcycle and bus using NTS-derived shares for the record’s region and trip-length band.

The shares sum to one within each cell, so the split conserves the parent volume; cells without NTS road evidence default to private car. The result is seven output modes: walking, cycle, private car, motorcycle, bus, rail and subway. Purpose is not recorded in the mobile input at all, so it is estimated entirely from survey and census evidence.

The estimate rests on trip-production rates: the average number of activities a person in a population segment undertakes for a purpose, mode and time period. Multiplying those rates by the population resident in a zone gives the activities that zone would be expected to generate for each purpose.

For each origin, resident population segments are weighted by their trip-production rates to give a local prior over purposes. The local prior determines how one origin’s trips divide between purposes, but it does not determine the national picture: summed across the country, the purpose mix is whatever the trip-rate tables imply.

An optional final stage anchors that mix to published survey controls while allowing each origin to differ from the national average according to who lives there. The control table reports trip start time by purpose for England from Monday to Friday, using the 2023–24 survey year.

Its weekday-only scope means weekday periods are controlled, while weekend travel is left to the local prior. Passively generated datasets lack a ground truth, and there is no independent small-area OD measurement for England and Wales against which these matrices could be checked.

So the evaluation provides consistency checks rather than a verdict on accuracy. The checks compare the results with survey evidence and ask whether the result behaves as intended. The pipeline regenerates these checks on every run, so rebuilding the dataset under different assumptions produces equivalent evidence for the new output.

The mode-share comparison covers 427 region–band–mode cells with an NTS target: all the English cells, with the remaining 25 being Welsh and untargeted. It uses the seven output modes after calibration and the road-mode split.

Across those cells, the mean absolute gap falls from 5.07 to 0.01 percentage points, or from 7.86 to 0.01 when each cell is weighted by volume. The private-car gap falls from 13.5 to 0.03 percentage points, while the walking gap falls from 11.6 to below 0.01.

In aggregate, calibration moves the base from 88.8 percent road and 8.1 percent walking to NTS reference shares of 65.4 percent and 31.0 percent. Only one cell keeps a gap above 0.5 points, because the survey reports a mode the mobile data never recorded, and calibration cannot create travel the source does not contain.

Raking moves the weekday morning-peak purpose mix substantially. Before raking, the local prior assigns 35.7 percent of morning-peak volume to commuting, compared with an NTS0502 share of 20.4 percent. Personal business and holiday travel begin at 7.3 percent and 2.4 percent, against NTS0502 shares of 15.6 percent and 10.2 percent.

After raking, every control group matches its NTS0502 share to within 0.01 percentage points, while purpose composition still varies between origins with the local prior. Purpose is inferred rather than observed, so users whose questions turn on purpose should inspect the trip-rate priors and controls against local evidence.

In panel a, the largest visual adjustment is for walking: its blue bar sits near eight percent, while the calibrated bar and black target align around thirty-one percent. Private-car shares move toward the target as well, whereas bus, rail, and subway change only modestly.

Because the matrices are segmented by mode and period on one national geography, and by purpose for the weekday morning peak, analysts can extract the flows a question needs rather than using an undifferentiated total. Examples include morning-peak education trips for school-place and access studies, morning-peak car flows for corridor and congestion work, and bus and rail volumes for public-transport appraisal.

The product is synthetic: a planning estimate, not a measurement of individual travel. Calibration can act only on the dimensions it compares, so biases that do not appear in those margins pass through. Mobile network data are not representative: penetration varies between operators, and frequent device users are over-represented and more mobile than average, which tends to overstate mobility levels.

The NTS has covered England only since 2013, so mode-share calibration reaches no Welsh origin. Welsh mode mix is the mobile operator’s classification rather than a survey-referenced one. Child travel is added by uplifting adult volumes at the origin, so it inherits the adult OD pattern; independent school travel and school-destination structure may therefore be understated.

Purpose is inferred rather than observed, and sampling error in the NTS controls transfers into the product. The configurable, auditable pipeline lets users change an assumption and rebuild. The result is a dense, openly documented set of mode- and purpose-specific travel flows, but it is a synthetic planning estimate, not ground truth.

Its value lies in combining scale with inspectable assumptions—and making those assumptions changeable.