Change Detection at Scale: A Webinar Recap

We are excited to share the recording for makepath’s webinar, “Detect Change with Satellite Imagery and Open Source Machine Learning”.

We hosted the webinar alongside makepath partners, Planet (satellite imagery), Regrid (parcel data) and NVIDIA (advanced GPU tools). Our work focused on predicting future development in the Austin, Texas metro area by using Machine Learning methods on satellite imagery.

You can watch the webinar in its entirety below.

Why Should You Care?

The main use case we foresee with change detection analysis is the ability to predict future land usage and invest in high-potential areas early on.

The goal is to be able to detect changes as it applies to real estate investment opportunities. So, end users will see where new parcels are popping up that have not been developed yet. 

Exploring Change Detection Across Central Texas

Dr. Dylan Stewart, Senior Machine Learning Engineer at makepath, kicked off the event with an overview of a change detection analysis of the Greater Austin area.

A mosaic of Planet imagery from 2017 (in gray) with changes detected in 2022 (in green).

A key finding the Open Source machine learning model used was not trained to detect but nonetheless detected are Sometimes Islands in the middle of Lake Travis, visible in 2022 but not in 2017 (in the images below). Sometimes islands appear during periods of drought or low water levels.

The water level at Lake Travis, a portion of the Colorado River west of Austin, in 2017. Satellite imagery: Planet Labs PBC. Green portions indicate changes detected in 2022.

The water level at Lake Travis, a portion of the Colorado River west of Austin, in 2022. Satellite imagery: Planet Labs PBC.

Dr. Stewart also explored neighborhood development patterns over the last 5 years.

In looking at the grayscale image below from 2017, we can see areas that have been developed between 2017 and 2022 (in green). 

2017 raster of a neighborhood in Pflugerville, Texas. Satellite imagery: Planet Labs PBC.

The image below shows the actual changes that took place in the neighborhood, i.e. increased roads and real estate development.

2022 raster of a neighborhood in Pflugerville, Texas. Satellite imagery: Planet Labs PBC.

To learn more about the technical specifics behind our change detection analysis, check out our technical deep dive blog post here.

Planet: Mapping the World Daily through Satellite Imagery

Planet currently has the largest constellation of earth observation satellites, which made them a perfect partner for this project. 

With access to Planet imagery or Open Source GIS tools such as OpenStreetMap (a geographic database of the whole world), users can track urban development across areas of interest across the globe.

The example below tracks urban development in Tanzania using satellite imagery from 2018.

Basemap of Dar-es-Salaam, Tanzania, 2018.

The images below highlight a comparison of Planet building footprints and OpenStreetMap building footprints. This underscores that public imagery and datasets may have limitations sometimes due to the time it can take to update such data.

Planet change detection target area.

Comparison of Planet building footprints vs OpenStreetMap building footprints.

Regrid: Making Open Data Usable

Regrid is a property data and mapping company that collects data on every property in the U.S. 

Regrid Parcel data offerings.

makepath partnered with Regrid to get access to open parcel data that is standardized, clean and enhanced. Regrid offers useful additions for parcel data such as digitized parcels for municipalities where digitized data does not exist, custom data formats and flexibility in working with different tech stacks.

Parcel data use cases.

NVIDIA: Speeding Up and Powering Change Detection Analysis

makepath and NVIDIA have enjoyed a partnership that centers around mutual collaboration. makepath brings expertise in developing Open Source tools and pushes key Open Source libraries and feature development. NVIDIA brings a wealth of expertise and resources for advanced GPU capabilities. 

NVIDIA is increasingly playing a role in accelerating scientific computing from days to seconds, which is a huge asset for large-scale change detection analysis.  

What’s next?

makepath is working on a limited release beta change detection product in the coming year. To learn more and to stay up to date with the product roadmap, visit makepath.com/change. 

If you have any further questions, connect with us at contact@makepath.com or tweet us at @makepathGIS.