Analyzing origin and destination patterns
Last updated
Last updated
This tutorial leverages the Spatial Index H3 to visualize origin and destination trip patterns in a clear, digestible way. We'll be transforming 2.5 million origin and destination locations into one H3 frequency grid, allowing us to easily compare the spatial distribution of pick up and drop off locations. This kind of analysis is crucial for resource planning in any industry where you expect your origins to have a different geography to your destinations.
You can use any table which contains origin and destination data - we'll be using the NYC Taxi Rides demo table which you can find in the CARTO Data Warehouse (BigQuery) or the CARTO Academy Data listing on the Snowflake Marketplace.
In the CARTO Workspace, head to Workflows and Create a Workflow, using the connection where your data is stored.
Under Sources, locate NYC Taxi Rides (or whichever input dataset you're using) and drag it onto the workflow canvas).
When running origin-destination analysis, it's important to think about not only spatial but temporal patterns. We can expect to see different trends at different times of the day and we don't want to miss any nuances here.
Connect NYC Taxi Rides to a Spatial Filter component.
Set the filter condition to PART_OF_DAY = morning (see screenshot above). You can pick any time period you'd like; if you select the NYC Taxi Rides source, open the Data preview and view Column Stats (histogram icon) for the PART_OF_DAY variable, you can preview all of the available time periods.
Note we've started grouping sections of the workflow together with annotation boxes to help keep things organized.
The 2.5 million trips - totalling 5 million origin and destination geometries - is a huge amount of data to work with, so let's get it converted to a Spatial Index to make it easier to work with! We'll be applying the straightforward approach from the Convert points to a Spatial Index tutorial.
Connect the match output of the Simple Filter to a H3 from GeoPoint component and change the points column to PICKUP_GEOM; which will create a H3 cell for each input geometry. We're looking for junction and street level insights here, so change the resolution to 11.
Connect the output of this to a Group by component. Set the Group by column to H3 and the aggregation column to H3 (COUNT). This will count the number of duplicate H3 IDs, i.e. the number of points which fall within each cell.
Repeat steps 1 & 2, this time setting the initial points column to DROPOFF_GEOM.
Add a Join component and connect the results of your two Group by components to this. Set the join type to Full Outer; this will retain all cells, even where they don't match (so we will retain a H3 cell that has pickups, but no dropoffs - for instance).
Now we have a H3 grid with count columns for the number of pick ups and drop offs, but if you look in the data preview, things are getting a little messy - so let's clean them up!
Create Column: at the moment our H3 index IDs are contained in two separate columns; H3 and H3_JOINED. We want just one single column containing all IDS, so let's create a column called H3_FULL and use the following CASE statement to combine the two: CASE WHEN H3 IS NULL THEN H3_JOINED ELSE H3 END.
Drop Columns: now we can drop both H3 and H3_JOINED to avoid any confusion.
Rename Column: now, let's rename H3_COUNT as pickup_count and H3_COUNT_JOINED as dropoff_count to keep things clear.
Now, you should have a table with the fields H3_FULL, pickup_count and dropoff_count, just like in the preview above!
Now, we can compare the spatial distribution of pickups and dropoffs:
Connect two subsequent Normalize components, first normalizing pickup_count, and then dropoff_count. This will convert the raw counts into scores from 0 to 1, making a relative comparison possible.
Add a Create Column component, and calculate the difference between the two normalized fields (pickup_count_norm - dropoff_count_norm). The result of this will be a score ranging from -1 (relatively more dropoffs) to 1 (relatively more pickups).
You can see the full workflow below.
Check out the results below!
Do you notice any patterns here? We can see more drop offs in the business district of Midtown - particularly along Park Avenue - and more pick ups in the more residential areas such as the Upper East and West Side, clearly reflecting the morning commute!
