Predict
Estimate hourly pickup demand for every mapped NYC TLC zone at a selected forecast horizon.
NYC urban demand intelligence • AWS SageMaker AI
Forecast where NYC demand moves next.
Flowcast predicts hourly taxi pickup demand across New York City at 1-, 6-, and 24-hour horizons. It combines historical TLC trips, weather, calendar patterns, and zone context, then serves the forecasts through AWS SageMaker AI.
- NYC-wide zone coverage · 262 zones
- 1h · 6h · 24h horizons
- XGBoost + SageMaker AI
What it does
From raw city activity to a forecast you can actually read.
Understand
See where demand is concentrated, how zones compare with the one-week baseline, and how uncertainty changes across the city.
Validate
Replay held-out historical periods to compare predictions with observed demand and measure where the model succeeds or misses.
How to use Flowcast
Three steps from citywide signal to zone-level detail.
- 1
Choose a time
Select a forecast origin and a 1h, 6h, or 24h horizon.
- 2
Scan the city
The map colors every available NYC taxi zone by predicted hourly pickups.
- 3
Inspect a zone
Hover or click for the forecast, uncertainty, baseline comparison, and local trend.
Why spatial demand matters
Demand is not only a number. It has a place and a time.
A citywide total cannot show where the next busy hour will happen. Flowcast turns demand into a spatial, time-aware signal that can be explored zone by zone across New York City.
Potential planning applications. Flowcast is a portfolio project and is not currently used by any operator.
Fleet and dispatch planning
Anticipate which zones are likely to need more vehicles an hour, a shift or a day ahead.
Resource allocation
Compare expected demand across boroughs before deciding where to place capacity.
Operations and staffing
Plan staffing around the hours and places where demand is expected to peak.
Urban demand analysis
Study how time, weather and geography shape mobility demand across the city.
Forecast horizons
One forecasting system. Three planning windows.
A single global XGBoost model serves every zone and every horizon; the horizon is one of its inputs, alongside recent demand, calendar, weather and location.
1 HOUR
Near-term operational demand by zone.
6 HOURS
Same-day demand outlook for planning ahead.
24 HOURS
Next-day zone-level demand horizon.
Product preview
Every zone, one glance.
This is the real next-hour forecast from the latest published data. Brighter zones expect more pickups. The Forecast page adds time controls, uncertainty, baseline comparisons and zone-level detail.
Predicted pickups / hour
< 11–55–1515–4040–100100–250≥ 250
Lower demandHigher demand
Model proof
A forecast should be measurable.
Flowcast evaluates held-out periods against the same zone and hour one week earlier. Published model metrics must come from generated evaluation artifacts, never hard-coded values.
WAPE
18.6%
Baseline 24.7%
MAE
3.72
pickups per zone-hour · baseline 4.94
RMSE
11.52
baseline 15.63
Baseline improvement
+24.8%
relative WAPE reduction
Held-out test period Mar 1, 2026 – Jul 31, 2026, 262 zones, all three horizons. Model rsx1twqvdz9b.
AWS architecture
Built beyond the notebook.
Flowcast connects data processing, model training, validation, model registration, inference, APIs, and an interactive geospatial product in one end-to-end system.
01
Data
02
Features & training
03
Validation & registry
04
Inference & product