01 · OBSERVE
INTERPRETABLE URBAN DATA STUDY · CENTRAL MELBOURNE · 2025
When does a place depart
from its usual rhythm?
This project compares 2025 hourly pedestrian counts from 12 purposively selected central-Melbourne sensors with each location's own weekday–hour baseline. Strong consecutive departures become Episodes; same-direction co-occurrence across at least three sensors becomes a Pulse. Under the v1 definitions, the workflow produced 425 Pulses, with 16 cases selected for manual evidence review.
Reviewed context records overlap with a signal; it does not establish its cause.
SELECTED PULSE · NEW YEAR, PHASE 1Above local baselines across the research network1 JAN 2025 · 00:00–05:00 · 9–12 ACTIVE SENSORS PER HOUR · 12 INVOLVED
LOADING VERIFIED PULSE DATA
Discrete sensor results · no interpolated city surface▲ ABOVE BASELINE · ● SELECTED PULSE06:00–10:00 → below-baseline phase
- Central-city research sensors
- 12
- Pulses under v1 rules
- 425
- Manually reviewed cases
- 16
SCROLL TO FOLLOW THE STUDYDetection first · context reviewed afterwards
01 · RESEARCH QUESTION
When does a place depart from its own usual rhythm?
A pedestrian count has little meaning without a local reference. The same value may be ordinary at one place and unusual at another, or ordinary on a weekday morning and unusual late at night. This study asks how far an hourly observation departs from the usual distribution for the same sensor, weekday and hour; how long that departure persists; and whether the same direction appears across several research locations.
LOCAL BASELINE → HOURLY DEPARTURE → EPISODE → PULSE
PULSE WINDOW
A Pulse is a rule-defined period in which at least three research sensors show strong departures in the same direction during every included hour. It describes temporal co-occurrence across geolocated sensors; the v1 grouping does not use distance or spatial adjacency.
02 · METHOD IN BRIEF
From hourly observations to reviewed Pulses.
The workflow keeps measurement, baseline construction, deviation detection, cross-location grouping and evidence review separate. The rules remain visible so that each result can be traced back to the observations and decisions that produced it.
02 · BASELINE
Local reference
SENSOR × WEEKDAY × HOUR · MEDIAN · RAW MAD03 · DETECT
Consecutive departures
HIGH-CONFIDENCE BASELINE · |SCORE| ≥ 3 · SAME DIRECTION04 · GROUP + REVIEW
Cross-location Pulse
≥3 SENSORS PER HOUR · CONTEXT REVIEWED AFTER DETECTIONThe score is the observed count minus the local baseline median, divided by the raw MAD. It is a relative deviation measure, not a probability.Read the full methodology →
03 · CASE 01 / NEW YEAR
One date, two directions.
Across the selected research network, a six-hour above-baseline Pulse from 00:00 to 05:00 was followed by a five-hour below-baseline Pulse from 06:00 to 10:00. The hourly number of participating sensors changed throughout both periods; the result is a direction change with an evolving participation pattern.
PHASE 1 · ABOVE BASELINE
00:00–05:00
6 HOURS · 9–12 ACTIVE SENSORS PER HOUR · 12 INVOLVED · NETWORK-WIDE AT PEAK
Reviewed context: New Year events, early-morning crowds, extended public transport and road closures
Inspect this Pulse →PHASE 2 · BELOW BASELINE
06:00–10:00
5 HOURS · 7–12 ACTIVE SENSORS PER HOUR · 12 INVOLVED · NETWORK-WIDE AT PEAK
Reviewed context: public-holiday morning, business closures and the end of overnight transport changes
Inspect this Pulse →The reviewed sources overlap different parts of the sequence. They help describe its context but do not identify one cause for either phase.
04 · CASE 02 / WET WEATHER
Two wet periods, two temporal structures.
Both selected Pulses are below baseline and overlap recorded rain. One lasts three morning hours; the other continues for nine hours from midday. The comparison describes different durations and participation patterns within two purposively selected cases. It does not estimate a general rainfall response.
6 JAN · WET MORNING
07:00–09:00
3 HOURS · 6–9 ACTIVE SENSORS PER HOUR · 10 INVOLVED
3.6 mm rain across 3 hours · school holiday · Australian Open qualifying in the wider context
Inspect this Pulse →2 JUL · SUSTAINED WET PERIOD
12:00–20:00
9 HOURS · 3–11 ACTIVE SENSORS PER HOUR · 12 INVOLVED
10.4 mm rain across 8 rainy hours · cold conditions · limited later transport context
Inspect this Pulse →The two reviewed wet-weather cases differ in duration and hourly sensor participation. Because they were selected for comparison and other conditions were not controlled, the study treats this as a descriptive contrast rather than a causal weather effect.
05 · CASE 03 / ACTIVITY CONTEXT
Event-overlapping signals do not share one form.
The selected Melbourne Marathon Pulse is concentrated in one early-morning hour. The 17 December case extends across eight evening and overnight hours. Both are above baseline and overlap documented activities, but their durations and participation structures differ.
MELBOURNE MARATHON · 06:00
One-hour Pulse
1 HOUR · 9 ACTIVE SENSORS · 9 INVOLVED · ABOVE BASELINE
Reviewed context: marathon access, road closures and public-transport arrangements
Inspect this Pulse →17 DEC · 18:00–01:00
Eight-hour evening Pulse
8 HOURS · 3–8 ACTIVE SENSORS PER HOUR · 10 INVOLVED · ABOVE BASELINE
Reviewed context: RMIT graduation and two Christmas activities with partial time overlap
Inspect this Pulse →Within this purposively selected case pair, documented activity context overlaps two substantially different temporal forms. The comparison does not estimate attendance-normalised effects or event causation.
06 · CASE 04 / UNRESOLVED
A detected signal can remain unexplained.
An eight-hour above-baseline Episode at QVM–Therry Street South remained confined to one sensor. No matching night event was identified within the reviewed sources. The result is retained as unresolved because source coverage cannot prove that no local circumstance occurred.
QVM–Therry St (South)20:00–03:00 · 8 HOURS · ONE SENSOR
ISOLATED ABOVE-BASELINE EPISODE
EVIDENCE REVIEW
No matching night event identified in reviewed sources
- Regular market hours
- NO TEMPORAL MATCH
- Night-event search
- NO MATCHING SOURCE IDENTIFIED
- Direction assessment
- INDETERMINATE
LIMITED NEGATIVE SEARCH
Not evidence that no local circumstance occurred.
Unresolved signals remain visible so that a missing explanation is not replaced by an invented one.
07 · CURRENT SCOPE + FURTHER STUDY
What the v1 framework establishes.
The current study provides a reproducible result for a defined 2025 sensor network and transparent rule set. Its boundaries describe how the findings should be read and where later research could add resolution.
ESTABLISHED IN V1
What can be inspected
- Direction relative to each local weekday–hour baseline
- Duration and hourly sensor participation
- Cross-location co-occurrence under explicit thresholds
- Manually reviewed context overlap and unresolved cases
INTERPRETATION BOUNDARY
How to read the results
- A central-city study network, not a city-wide statistical sample
- Peak participation categories, not spatial clustering
- Purposefully selected cases, not a representative sample of all Pulses
- Context overlap, not an estimate of causal effect
FURTHER STUDY
Ways to add resolution
- Season- and event-aware baselines
- Distance-, precinct- or topology-aware grouping
- Threshold calibration with annotated or held-out data
- Larger sensor networks and causal study designs
08 · EXPLORE THE 2025 RESULTS
Inspect the complete
Pulse set.
The case studies show selected contrasts. The Explorer exposes all 425 v1 Pulses across the 12 research sensors, including filters, direction, scope, hourly participation, evidence status and sensor-level context.
Open the New Year caseINTERACTIVE 2025 RESULT EXPLORER
09 · RESEARCH PROCESS
Human-directed research,
developed with AI assistance.
I defined the research topic, designed the project and interaction architecture, directed each stage of development, and reviewed outputs against my intended research and visual direction. AI tools assisted with analytical implementation, code, source discovery, documentation and testing. The published results come from deterministic scripts and explicit rules rather than an AI prediction model.
Research direction + design
Topic, project architecture, interaction design and final direction · author-led
Analytical + software implementation
Methods, data processing and interface development · AI-assisted, author-directed
Evidence review
Public-source review, case interpretation and uncertainty decisions · author-directed
Verification + editorial control
Automated consistency tests, iterative review and final approval
AI supported the research process; it was not the analytical model and did not independently determine causal explanations.