Machine Learning for Change Detection: Part 3

This blog post is part three of a four-part series, “Machine Learning for Change Detection”.

In part one, we made a change detection map for an area of 6,750 sq. mi. using Planet’s satellite imagery.

In part two, we zoomed in and used time sliders to visualize changes to specific areas.

For part three, let’s predict the future development of Austin by comparing our change maps with Regrid’s parcel data.

By finding where parcel data exists but no change has occurred yet, we can predict future development in the Austin area.

Prediction map for future development of Austin, Texas.

Satellite imagery: Planet Labs PBC.

Green areas highlight parcels we detected with no changes, which means potential for future development in those areas.

This follows the growth trends we saw in part one, where development was concentrated along the north-south I-35 corridor.

Coupland, Texas: 2017, 2022, and change detection map with Regrid Parcels.

Satellite imagery: Planet Labs PBC. [Full Screen]

In Coupland, these parcels are clearly divided— but there is no significant land cover change between 2017 and 2022.

Takeaway 1: Parcels with no change detected suggest future development in that area.

Takeaway 2: Parcels with no change could be potential investment opportunities for inquiring developers.

Pflugerville: 2017, 2022, and change detection map with Regrid Parcels.

Satellite imagery: Planet Labs PBC. [Full Screen]

In Pflugerville, parcels set aside in 2017 transformed into a large neighborhood by 2022.

Takeaway 1: The parcel data reflects development planning at or before 2017.

Takeaway 2: The green change detection map illustrates the delivery of those plans over the span of 5 years.

Round Rock, Texas: 2017, 2022, and change detection map with Regrid Parcels.

Satellite imagery: Planet Labs PBC. [Full Screen]

In Round Rock, urban development sprawled along an adjacent parcel.

Takeaway 1: This area could be parceled out to different developers.

Takeaway 2: Development not only occurs within designated parcels, but directly around parcels as well.

Our change detection maps and parcel maps are correlated.

Just like in parts one and two, we leveraged Open Source Machine Learning tools and Open Source GIS tools to complete our analysis.

These tools combined with GPU-enhanced analytics on a NVIDIA A100 GPU allows us to classify large amounts of data in minutes.

Do you want to know the future development hot spots in your local area?

What would you do if you knew where future development would take place?

Let us know on Twitter @makepathGIS, or email us at contact@makepath.com.

Popular