Automating the creation of custom land use map at scale

Company description

Advanced Environmental Services (AES) is a leading company in Environmental Services in the United States. They provide advice, analysis and operational support services for natural resources management for every type of customer, all over the country. 

In a lot of their urbanism and forestry projects, they need to generate land use maps with a high level of detail. This is why they decided to work with SpaceSense.

The challenge

In most of the projects that AES does, they need to create a land use map at some point. Depending on the project, they have to use a different level of spatial granularity (they used NAIP, Sentinel and Maxar) and a different number of types of land use to track.  Furthermore, they need to do that all over the United States, with major geospatial differences between each state. 

This means that they cannot re-use what they build in previous projects, and that they have to start from scratch most of the time. Additionally, they are building these maps almost manually. As expected, this is very time consuming and represents a significant cost for AES.

They want to accelerate the creation of these maps through machine learning applied to satellite and aerial imagery. Since they do not have the in-house expertise, they decided to work with SpaceSense

The solution

‍

AES used SpaceSense’s land use “Solution Template”. These “Solution Template” are pre-written workflows designed to train specific models, in this case a model to build land use maps. The user only has to integrate their custom training data, and the template does the rest automatically by generating a custom trained model. The user can then tweak several training parameters to improve performance if it is necessary.

The template has the following steps:

1. Gather the imagery type for which you want to build the land use map, with a small sample of annotation layer - 3 days
2. Customize the “Solution Template” for land use by incorporating the data gathered in step 1, and validating all training parameters - 2 day
3. Train and test the land use model. There might be several iterative loops where you change some parameters to improve the accuracy of the model - 2 days
4. Run the model over the region of interest to get results on demand - 1 day

‍

The result

Thanks to SpaceSense, AES is able to build a new land use map for each project with an accuracy superior to 90% for all the monitored classes in less than a week.

They estimate that on average, they save about 85% of their time compared to the previous method, without sacrificing the quality of the maps.